Common beanAerial / UAVField / plotMultispectral / hyperspectralWhole plant / canopy / plot / fieldClassificationGrowth / development / phenology
Accurate and scalable prediction of physiological maturity (PM) in leguminous crops remains a key challenge due to indeterminate growth and canopy heterogeneity. Thus, causing difficulties for optimizing breeding decisions and field management. This study develops a novel UAV-based multispectral-index fusion framework for high-throughput maturity classification in dry bean. This is conducted by combining parametric and non-parametric machine learning (ML) classification models to capture non-linear maturity signatures. Multispectral imagery was collected over three growing seasons across multiple genotypes. Six spectral bands and five maturity-relevant vegetation indices capturing chlorophyll degradation, senescence, and canopy greenness were evaluated through 63 feature-set combinations and 10 parametric and non-parametric ML classifiers. Model performance was assessed using stratified five-fold cross-validation, composite z-score aggregation, and Friedman-Nemenyi statistical ranking to jointly evaluate accuracy and consistency. Among all the model-feature combinations, the Support Vector Classifier (SVC) paired with Band_All6 + MCARI feature-set achieved the highest test accuracy (>70%), with class-wise recall of 63.2%, 71.4% and 79.4% for early, medium and late maturity, respectively. Hybrid feature-sets integrating chlorophyll-sensitive indices with spectral bands consistently outperformed index-only or band-only sets, confirming the synergistic effect of bands and engineered index coupling. This research establishes a statistically validated pipeline linking UAV multispectral data, vegetation spectral-index and machine learning classification to quantify PM variability in dry bean with potential for validation in other legume crops. The proposed framework offers a generalizable approach for non-destructive maturity prediction, enabling breeders to accelerate genotype selection and harvest scheduling under field-scale conditions.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像と機械学習による乾燥豆の生理的成熟度推定パイプラインを開発・統計検証しており、植物状態の取得・抽出手法が研究の中心である。
abstractThis study develops a novel UAV-based multispectral-index fusion framework for high-throughput maturity classification in dry bean.
Hyperspectral imaging (HSI) systems offer rich spectral information for precision agriculture applications such as yield forecasting, but their high cost and complexity limit widespread adoption. The relationship spectral data complexity and prediction accuracy in non-destructive crop monitoring remains unclear, challenging the assumption that more complex spectral data inherently yields better predictions. We developed a workflow integrating UAV-based HSI (150 bands) with deep learning to non-destructively estimate fresh weight of cabbage heads (n = 680). A two-dimensional convolutional neural network (2D-CNN) was deployed to identify the single most predictive wavelength through systematic feature extraction. Performance was benchmarked against 20 conventional multi-band vegetation indices (VIs). The CNN identified a single spectral band at 565.63 nm as the optimal predictor. A predictive model using only this single-band input achieved a coefficient of determination (R²) of 0.49 on an independent test set, with a root mean square error (RMSE) of 0.99 kg and a mean absolute error (MAE) of 0.84 kg. This performance substantially surpassed the best-performing conventional multi-band VI (MCARI2, R² = 0.34), representing a 44% improvement in explained variance using single-band data input. This AI-driven approach autonomously distill HSI complexity into a single optimal wavelength that outperforms established multi-band indices. The findings provide a methodological framework for designing simplified, cost-effective spectral sensors for precision agriculture, potentially improving accessibility and scalability of crop monitoring technologies.
Why it matches plant phenotyping methodsキャベツ結球の生体重という植物形質を、UAVハイパースペクトル画像とCNNで非破壊推定するワークフローを開発し、独立テストと既存指標との比較で検証しており、形質取得手法が中心である。
abstractWe developed a workflow integrating UAV-based HSI (150 bands) with deep learning to non-destructively estimate fresh weight of cabbage heads (n = 680).
Sugarcane is a high-value industrial crop vital for sugar and biofuel production, yet increasingly constrained by climate variability, biotic and abiotic stresses, soil degradation, and inefficient input use. Traditional breeding and crop management approaches are often slow, labour-intensive, and less precise, emphasizing the need for digital transformation in sugarcane agriculture. AI now offers powerful tools to accelerate genetic improvement, enhance stress resilience, and optimize resource-use efficiency. This review synthesizes recent advances in AI applications across the sugarcane improvement pipeline, including high-throughput phenotyping, genomic prediction, digital crop monitoring, and AI-driven decision-support systems. ML and DL models enable automated, accurate prediction of key traits such as biomass, canopy temperature, nitrogen status, and sugar recovery using UAV, satellite, and proximal sensing data. AI-powered genomic selection approaches leveraging convolutional networks, transformers, and attention mechanisms improve prediction accuracy for yield, ratooning ability, and stress tolerance by integrating SNPs, pedigree, and multi-environment datasets. Emerging innovations such as digital twins, multimodal data fusion, reinforcement learning-based irrigation scheduling, and climate-smart advisory models further strengthen real-time crop intelligence. The integration of blockchain-enabled breeding databases, FAIR data standards, and interoperable analytics pipelines supports scalable and collaborative research. Literature analysis reveals 15-30% gains in selection efficiency, >90% accuracy in disease detection, and phenotyping cost reductions of up to 70%. Key challenges remain, including scarce annotated datasets, genotype × environment complexity, model interpretability, and adoption barriers for smallholders. A future roadmap is proposed featuring multimodal foundation models, edge-AI deployment, and explainable breeder dashboards. AI is redefining sugarcane research from reactive to predictive, enabling climate-resilient, sustainable, and profitable production systems.
Why it matches plant phenotyping methodsサトウキビ育種におけるAI応用の総説であり、高スループット表現型解析、UAV・衛星・近接センシングによる形質推定を主要な対象として扱っているため、フェノタイピング手法レビューとして適格。
abstractThis review synthesizes recent advances in AI applications across the sugarcane improvement pipeline, including high-throughput phenotyping, genomic prediction, digital crop monitoring, and AI-driven decision-support systems.
Accurate estimation of leaf chlorophyll content is essential for monitoring crop growth and supporting precision agricultural management. The soil and plant analyzer development (SPAD) instrument readings represent the relative chlorophyll content (RCC) in leaves, a key indicator of photosynthetic capacity and physiological status in wheat. This study proposes a multimodal data fusion approach integrating unmanned aerial vehicle (UAV)-derived vegetation indices (VI) and texture features (TF) from multispectral imagery with short-term environmental time-series data collected from in-field meteorological and soil sensors to estimate winter wheat RCC. A self-attention deep neural network (SA-DNN) was developed to capture complex nonlinear relationships among multimodal inputs. Employing a multi-stage progressive feature selection strategy that combines Pearson and Spearman correlations, minimum redundancy maximum relevance (mRMR), and least absolute shrinkage and selection operator (LASSO), eight optimal features were ultimately selected from VI and TF, together with indicators of short-term variability in environmental factors (EF). These features included OSAVI, NDRE, R-Mea, R-SEM, AH-std-7, DPT-std-7, SD-std-7, and pH-std-7. The SA-DNN model achieved the best estimation performance (R² = 0.913, RMSE = 3.945), significantly outperforming traditional machine learning models such as XGBoost, random forest (RF), support vector regression (SVR), Adaboost, and partial least squares regression (PLSR). SHapley Additive exPlanations (SHAP) analysis further quantified the contributions of individual features, revealing that short-term (7-day) fluctuations in EF, particularly sunshine duration and air humidity, played a dominant role in regulating variations in RCC. Overall, this study demonstrates that the synergistic integration of multimodal data within an attention-based deep learning framework significantly augments the precision of winter wheat RCC estimation, providing a powerful tool for real-time crop growth monitoring and the optimization of precision agricultural management.
Why it matches plant phenotyping methodsUAV画像・環境センサーデータから小麦葉のクロロフィル含量を推定する深層学習手法を開発し、複数モデルとの性能比較で検証しており、植物表現型取得が研究の中心である。
abstractThis study proposes a multimodal data fusion approach integrating unmanned aerial vehicle (UAV)-derived vegetation indices (VI) and texture features (TF) from multispectral imagery with short-term environmental time-series data collected from in-field meteorological and soil sensors to estimate winter wheat RCC.
Accurate and non-destructive estimation of rice Leaf Area Index (LAI) is vital for crop growth assessment and yield prediction. Close-range, non-contact optical methods are commonly used for LAI monitoring. However, their accuracy is often limited by the platform, and most rely on single-source data prone to saturation effects and background interference. To overcome these limitations, this study develops a phenotyping robot-based multi-source, high-resolution data fusion approach for field-scale LAI monitoring. A phenotyping robot equipped with multispectral and high-resolution RGB cameras was used to collect vegetation indices, color indices, texture features, and canopy coverage extracted from high-resolution RGB imagery. These features were further combined with meteorological variables to build machine learning models. The Random Forest model achieved the best performance (R² = 0.92, RMSE = 0.302). SHAP (Shapley Additive Explanations) was applied to interpret the model and quantify the importance of multispectral features, RGB-derived texture information, canopy coverage and meteorological factors. Canopy coverage, NDVI and Clgreen were identified as the key factors for improving model performance, and the complementary mechanism between canopy coverage and other features can alleviate the saturation effect in the high LAI stage. The results show that combining high-resolution remote sensing data from robots with meteorological data can effectively mitigate the saturation effect and soil background interference in LAI estimation, and significantly improve the accuracy of LAI estimation. This study provides a practical and scalable framework for field phenotyping and offers technical support for precise rice cultivation and smart agriculture.
Why it matches plant phenotyping methodsロボット搭載マルチセンサーと画像特徴量融合によるイネLAI推定手法を開発・評価しており、表現型取得・推定が研究の中心である。
abstractthis study develops a phenotyping robot-based multi-source, high-resolution data fusion approach for field-scale LAI monitoring.
Given the substantial agronomic and economic significance of maize, the development of real-time and high-precision disease detection methodologies is essential for ensuring yield stability. While hyperspectral imaging excels at capturing fine-grained spectral signatures of infection, its higher detection precision comes with considerable high hardware and temporal costs compared to RGB imaging, posing significant challenges for scalable field applications. To bridge this gap, this article proposes a fusion perception unfolding network (FPUF-Net) for high-fidelity maize spectral reconstruction from RGB images. Distinct from conventional deep learning models, FPUF-Net unfolds the optimization problem via a half-quadratic splitting algorithm, solving the data subproblem and prior subproblem alternately during iterations. Specifically, a fusion feature learning network and a spectral-spatial joint attention network are designed within the data subproblem to explicitly exploit RGB spatial priors and mitigate spatial smoothing. Moreover, a spectral-spatial Transformer is utilized as the denoiser to capture long-range spectral dependencies in the prior subproblem. Experiments performed on a maize spectral recovery dataset comprehensively demonstrate that FPUF-Net can effectively reconstruct maize hyperspectral images with superior precision. The structural characteristics enable the network to perceive long-term spectral-spatial fusion features, significantly reducing reconstruction errors, particularly in the biologically critical red-edge region (620-700 nm). In downstream disease detection tasks, the overall accuracy of reconstructed HSIs improves over RGB by margins of 0.51% to 8.1% across different scenarios, while the average accuracy increases by 2.45% to 18.86%. These results indicate that the proposed model offers a viable, cost-effective solution for applying hyperspectral imaging in field settings, enabling its scalable use in agricultural robots.
Why it matches plant phenotyping methodsRGB画像からトウモロコシのハイパースペクトル情報を再構成する手法を開発し、データセットで性能評価している。植物のスペクトル状態および病害検出に直接関わる手法が研究の中心である。
abstractthis article proposes a fusion perception unfolding network (FPUF-Net) for high-fidelity maize spectral reconstruction from RGB images.
The diameter of natural rubber trees serves as a critical crop parameter, not only for determining whether rubber trees meet tapping requirements and assessing their growth status, but also playing a significant role in calculating parameters such as tapping angle, trajectory, depth and yield prediction. To further advance the intelligent production level of natural rubber trees and achieve low-cost, automated diameter measurement methods, this study proposes a new approach that combines YOLO11s instance segmentation algorithms with a monocular RGB camera to enable non-contact and non-fixed distance diameter measurement of natural rubber trees. The YOLO11-seg is used to obtain masks and bounding boxes for the ID, trunk, and tapped area. This method employs image processing techniques such as contour smoothing, trunk skeleton extraction, angle calculation, and localization. With using the tree ID tag as the primary dimensional reference and incorporating the segmentation contours of trunk categories, it achieves the measurement and calculation of rubber tree trunk diameter. The results demonstrated that among the compared instance segmentation models, the highest segmentation accuracy mAP50-95ˢᵉᵍ reached 0.934. The diameter estimation based on this segmentation and geometric correction process achieved a root mean square error (RMSE) of 2.85 cm and a mean absolute percentage error (MAPE) of 12.58 % under the original measurement conditions. After error compensation, the RMSE and MAPE decreased to 2.13 cm and 8.68 %, respectively. The proposed method can accurately measure the diameter of natural rubber trees, significantly reducing the hardware cost. It provides a new approach for measuring the diameter of natural rubber trees, and also provides both theoretical support and practical basis for the intelligent production and precision agriculture in natural rubber cultivation.
Why it matches plant phenotyping methodsゴム樹の幹径という植物形態形質を、RGB画像とインスタンスセグメンテーションで非接触・自動推定する手法を開発し、精度検証まで行っており、フェノタイピング手法が中心である。
abstractthis study proposes a new approach that combines YOLO11s instance segmentation algorithms with a monocular RGB camera to enable non-contact and non-fixed distance diameter measurement of natural rubber trees.
Empirical and physical models are widely used for monitoring equivalent water thickness (EWT) to adjust plant moisture management. However, model transferability to different times and locations, and insufficient training data remain the two key challenges of field spectroscopy analysis. Therefore, this study aims to construct a hybrid model, which combines the physical models optimized by Wasserstein Generative Adversarial Nets (WGAN) and empirical models for performing hyperparameter searches (the process of finding optimal model settings) to monitor the peanut EWT. Specifically, we develop a large spectral dataset consisting of field-measured data which including 246 peanut varieties in five peanut farms across China and synthetic datasets generated from the physical models optimized by WGAN. Furthermore, the PWLEH was constructed by hyperparameter tuning and pre-training which using synthetic datasets, and then fine-tuned by modular training with field data of peanut canopy water content. Comparing the model constructed with field data (R² = 0.5618, mean squared error (MSE) = 0.0725) and PROSAIL (a widely used canopy radiative transfer model) (R² = 0.7105, MSE = 0.0473), PWLEH achieved high accuracy in predicting peanut water content (R² = 0.7650, MSE = 0.0519). Unlike pure data-driven approaches, the new hybrid model incorporated radiative transfer knowledge and obtained higher predictive performance with fewer field data. This study demonstrates the potential of applying an optimized PROSAIL, hyperparameter search and modular training to improve the accuracy and transferability of the EWT prediction model, providing a new approach for sustainable agricultural management.
Why it matches plant phenotyping methods落花生のキャノピー分光データから等価含水厚（EWT）を推定するハイブリッドモデルを開発・評価しており、植物水分形質の取得・推定手法が中心である。
abstractTherefore, this study aims to construct a hybrid model, which combines the physical models optimized by Wasserstein Generative Adversarial Nets (WGAN) and empirical models for performing hyperparameter searches
TomatoStem / branchPhysiological trait estimationWater status / transpiration
Accurate sap flow measurements in small-diameter plant organs are essential for understanding water transport and source-sink dynamics, yet existing methods are limited by their temporal resolution, reduced sensitivity to low or reverse flow, and incompatibility with small organ dimensions. In this study, the ExoHeat sensor, a continuous-heating solution was developed for bidirectional sap flow measurements in small-diameter plant organs. Its performance was validated on tomato truss peduncles (Solanum lycopersicum L.). Zero-flow corrections accounting for ambient temperature and peduncle diameter ensured robust baseline adjustment, while gravimetric calibrations revealed a strong linear relationship between the sensor-measured temperature difference and sap flow rate up to 2 g h⁻¹, corresponding to a sap flux density of 5.8 10⁻³ cm³ cm⁻² s⁻¹. Whole-plant validation further demonstrated close agreement between ExoHeat-derived sap flow and gravimetric transpiration data. Anatomical imaging showed an asymmetrical distribution of xylem vessels in the tomato truss peduncle, underscoring the importance of correct sensor orientation. High-resolution measurements on ripening trusses successfully captured dynamic bidirectional flow patterns. The ExoHeat sensor thus provides a novel, high-temporal-resolution tool for accurate monitoring of sap flow in small-diameter organs, with promising applications in plant physiology, irrigation optimisation and stress detection.
Why it matches plant phenotyping methods小径植物器官の双方向樹液流を測定するセンサーを開発し、重力法による校正・検証を行った、植物生理状態の取得手法が中心の研究。
abstractIn this study, the ExoHeat sensor, a continuous-heating solution was developed for bidirectional sap flow measurements in small-diameter plant organs.
Maize leaf morphology is poorly investigated because quantifying maize leaf geometry is still an open question due to the complexity of the 3D curved shape. By utilization of geometric curves, maize leaf morphology can be effectively described parametrically and quantitatively. We divided maize leaf into three components: midrib, cross-section and blade contour. Each component is represented by parametric curves and controlled by a group of parameters. A 3D maize leaf model is generated by translation, rotation and scaling of the three components. We demonstrated the parametric maize leaf model allows the applications of leaf geometry analysis, leaf-level radiation capture simulation and dataset synthesis for phenotyping pipeline. The parametric maize leaf model is configurable, extensible and scalable, allowing it to be used in agricultural digital-twin and high-accuracy phenotyping. It also has potential to serve as a platform for maize biophysical and biomechanical studies. The code for 3D maize leaf model generation is available at https://github.com/xzcppm/parametric_maize_leaf.
Why it matches plant phenotyping methodsトウモロコシ葉の3D形態・幾何をパラメトリックにモデル化し、表現型解析用データ合成にも利用できる手法を開発しているため、植物表現型取得・解析手法が中心である。
abstractA 3D maize leaf model is generated by translation, rotation and scaling of the three components.
Solanum tuberosum (potato) is one of the most important global food crops relative to economic opportunities and food security. Potato Virus Y (Potyviridae, PVY), a detrimental plant pathogen propagated by insect vectors, negatively affects tuber yield and quality. This has forced industry stakeholders to adopt many different types of mitigation strategies including pesticide applications, manual field scouting, and potato seed certification programs. Despite these efforts, PVY continues to disrupt industry production regions resulting in significant economic losses due to the lack of robust diagnostic tools. Machine learning algorithms trained on remotely sensed spectral features show promise as a diagnostic tool for many plant diseases including PVY. This study proposes a novel Convolutional Neural Network (CNN) architecture to detect potato plant canopy regions of plants infected with PVY based on unmanned aerial system (UAS) hyperspectral pixel features comprised of bands matching the center wavelengths of nine spectral channels captured by the European Space Agency’s Sentinel 2 multispectral instrument. Accuracy and F1 metrics of 0.815 and 0.766 respectively were achieved on test data collected over multiple growing seasons and locations. Additionally, efforts were made to identify optimal combinations of spectral bands that are most beneficial for the CNN classifier by evaluating every possible combination of the nine spectral wavelengths in groups ranging from 3 to 9 channels. Results show that hyperspectral channels centered on 783 nm, 739 nm, and 560 nm are the most important features for the CNN architecture. Additionally, six hyperspectral features consisting of the three previously mentioned along with 665 nm, 704 nm, and 864 nm yielded the best results of all possible combinations achieving accuracy and F1 Score metrics of 0.833 and 0.791 respectively.
Why it matches plant phenotyping methodsCNNとUASハイパースペクトル画像を用いて、PVY感染植物のキャノピー状態を検出する手法を開発・評価しており、植物病害表現型の取得が中心である。
abstractThis study proposes a novel Convolutional Neural Network (CNN) architecture to detect potato plant canopy regions of plants infected with PVY based on unmanned aerial system (UAS) hyperspectral pixel features
Effective feature representation and heterogeneous fusion are essential for plant leaf recognition. However, existing methods have several limitations, such as insufficient comprehensiveness and distinctiveness in feature representation, as well as a lack of full consideration for the compatibility and complementarity in heterogeneous fusion. In the end, we propose a discriminative shape representation named the bag of multiscale curvature angle cuts (BMCAC) to capture fine curvature and spatial distribution characteristics, an advanced deep representation called the progressive salient deep representation (PSDR) to fully exploit deep convolutional features, and an effective fusion framework termed the K-weighted shape and deep feature fusion (KWFF) to aggregate the local context and global importance of heterogeneous features. Specifically, BMCAC is derived from the curvature angle cuts (CAC), multiscale analysis, and the bag of visual words (BoVW) model; PSDR is constructed by applying progressive downsampling and hierarchical pooling operations to deep convolutional features; and KWFF is developed by encoding neighboring information using homogeneous distance measures while incorporating globally weighted contributions from heterogeneous distance measures. Extensive experiments on four well-known benchmark leaf datasets demonstrate that the proposed shape and deep representations can efficiently extract leaf image features, and the fusion framework can effectively integrate heterogeneous features, outperforming state-of-the-art methods. The source code is available at https://github.com/Mumuxi1123/BMCAC_PSDR_KWFF.
Why it matches plant phenotyping methods植物葉画像から形状・深層特徴を抽出し、認識のために融合する新規手法を開発・評価しており、葉の画像ベース表現抽出が研究の中心である。
abstractwe propose a discriminative shape representation named the bag of multiscale curvature angle cuts (BMCAC) to capture fine curvature and spatial distribution characteristics, an advanced deep representation called the progressive salient deep representation (PSDR) to fully exploit deep convolutional features, and an effective fusion framework termed the K-weighted shape and deep feature fusion (KWFF)
Accurate detection of winter damage in turfgrass is essential for proactive management but remains difficult because early-stage injury is faint, irregular, and easily confused with background noise. These characteristics create two major challenges: limited availability of reliable training data and the need for a segmentation model that is highly sensitive to subtle features. To address the data limitation, this study employs a Conditional Deep Convolutional Generative Adversarial Network (cDCGAN) to generate synthetic, high-fidelity vegetation index (VI) maps. Compared with raw spectral bands, VIs are more robust to noise and enhance both dataset diversity and model generalization. To meet the segmentation challenge, we introduce a Transformer-based model with a novel Adaptive Attention Decoder (AAD), which dynamically refines feature representations to improve detection of low-contrast, spatially irregular damage. Field experiments conducted on golf courses in central Oregon, USA, from 2022 to 2023 demonstrate that the proposed pipeline outperforms other advanced deep learning models, achieving an mIoU of 82.47%, an accuracy of 97.85%, a recall of 85.62%, and an F1-score of 88.30%. Overall, this research presents a problem-driven framework that integrates targeted data augmentation with an improved segmentation architecture, offering a robust and accurate solution for early detection of winter damage in precision turfgrass management.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像から芝草の冬害をセグメンテーションするデータ拡張・深層学習手法を開発し、圃場で性能検証している。植物の病害状態の取得が研究の中心である。
abstractTo address the data limitation, this study employs a Conditional Deep Convolutional Generative Adversarial Network (cDCGAN) to generate synthetic, high-fidelity vegetation index (VI) maps.
Accurate and efficient detection of the pollination status of strawberry flowers is essential for intelligent pollination robots, as it directly affects the determination of optimal pollination timing and improves fruit set rates. However, the small size of strawberry anthers, their visual similarity, varied opening states, and complex field environments make pollination status detection highly formidable. To overcome these constraints, this paper presents a streamlined and resource-efficient detection approach (ELSF-DETR), built upon the Real-Time DEtection Transformer (RT-DETR) and specially refined for detecting densely packed and visually similar small objects in agricultural scenes. A lightweight LS-ResNet backbone is constructed to better capture small and densely clustered anther structures in strawberry flowers while reducing model complexity for improved deployment efficiency. In addition, the integration of a P2 detection head with full-kernel convolution enhance the network’s capacity to focus on delicate anther contours and cracking characteristics. Furthermore, the Hierarchical Attention Fusion Block (HAFB) is employed to balance local detail extraction with global context understanding, reducing misjudgments caused by misleading fine-grained features. Lastly, by employing the Wise-IoU (WIoU) loss mechanism, the model achieves improved sensitivity to minor positional discrepancies in visually similar anther objects. Experiments conducted on a self-built strawberry flower dataset demonstrate that ELSF-DETR achieves superior performance, it achieves 88.2 % accuracy, 85.8 % recall, 87.1 % mAP@50, and F1 score of 86.98 %. Relative to the baseline architecture, mAP@50 and F1 improved by 7.1 % and 4.33 %, respectively, while the model parameters and GFLOPs were reduced by 6.86 MB and 13.7 G, meeting the requirements of high precision and low complexity. This work provides practical support for intelligent pollination systems in precision agriculture.
Why it matches plant phenotyping methodsイチゴ花の受粉状態という植物状態を画像から推定する検出モデルを開発・評価しており、植物フェノタイピング手法が中心である。
abstractthis paper presents a streamlined and resource-efficient detection approach (ELSF-DETR)
The strip intercropping of soybean and maize, characterized by planting the two crops alternately in adjacent rows, has been widely promoted in several regions of China due to its potential to enhance resource utilization efficiency and overall yield. Accurate detection of crop rows and missing seedlings is essential for enabling precision field operations such as variable fertilization and targeted spraying. Monocular vision has emerged as a core sensing modality owing to its low cost and high resolution. However, the significant differences in row and plant spacing between maize and soybean, coupled with complex field conditions such as weed interference and uneven emergence, severely limit the effectiveness of traditional image processing techniques based on thresholding and geometric fitting. These methods struggle to accommodate the morphological variability of multiple crops, resulting in poor row detection precision and unreliable identification of missing seedlings. In recent years, deep learning has shown strong performance in crop row detection and object recognition tasks, particularly through multi-task networks that integrate segmentation and localization-related features. Nevertheless, most existing studies focus on single-crop scenarios and often neglect the integration of agronomic knowledge, thereby limiting their robustness and interpretability in real-world field environments. To address these issues, this study proposes a Multi-Task Geometric Regression Network for row extraction and missing seedling detection in maize-soybean intercropping systems, built upon an improved U-Net++ architecture and guided by agronomic priors. The proposed method simultaneously performs crop segmentation, row direction prediction, and generation of a missing seedling heatmap. The geometric features output by the network are subsequently processed using agronomic prior-informed post-processing and geometric fitting to finally achieve row extraction and missing seedling localization. Agronomic constraints, such as row spacing regularity, are embedded in the loss function as prior-informed regularization terms, which further enhance detection accuracy and robustness in intercropped fields. Experimental results demonstrate that the semantic segmentation achieves an average Intersection over Union (IoU) of 0.82, an F1-score of 0.86, and a pixel accuracy of 0.91. Row centerline detection attains an F1-score of 0.86 and a mean offset (MO) of 3.9 pixels. For missing seedling detection, the crop classification accuracy reaches 0.91, the average localization error (ALE) is only 2.5 pixels, and the composite detection score (CD-F1) is 0.89. Compared with single-task methods without agronomic priors, the proposed multi-task framework exhibits significant improvements in both stability and accuracy for row detection and missing seedling localization in intercropping scenarios. These results provide practical guidance for deploying intelligent visual systems in precision agriculture and intercropping management.
Why it matches plant phenotyping methods作物画像から畝構造と欠株状態を抽出するマルチタスク手法を開発し、精度評価も行っており、植物状態の取得・推定が研究の中心である。
abstractthis study proposes a Multi-Task Geometric Regression Network for row extraction and missing seedling detection in maize-soybean intercropping systems
Accurate yield forecasting is crucial in optimizing resource management and decision-making processes in agriculture, particularly in crops such as strawberries, which require precise predictions due to their rapid and continuous ripening cycles. This study introduces PheMuT, a novel phenology-informed, multi-modal time-series model that integrates visual and meteorological data streams to enhance strawberry yield forecasting. The proposed method employs advanced computer vision techniques, including two YOLOv11 detectors, an optimized ByteTrack tracker, Segment Anything (SAM), and Depth Anything v2 (DAv2), for precise fruit detection, canopy, and volume estimation. Concurrently, high-frequency weather data are processed using a self-supervised autoregressive Temporal Convolutional Network (TCN), resulting in concise and informative weather embeddings. These visual and weather features are fused within an LSTM-based model to produce weekly yield forecasts. PheMuT was validated using two strawberry cultivars at a Florida research facility over two consecutive seasons. Results indicated that PheMuT improved forecasting accuracy, reducing mean absolute error (MAE) by 10.7%, root mean squared error (RMSE) by 12.5%, and mean absolute percentage error (MAPE) by 18.6% compared to baseline manual methods. Additionally, the model exhibited a notable improvement of 17.2% in the coefficient of determination (R²). PheMuT offers an efficient, automated framework for yield forecasting. Code and data are available athttps://github.com/Sycamorers/PheMuT. The full datasets used in this study are available from the authors upon request.
Why it matches plant phenotyping methods果実検出、キャノピー・体積推定などの画像ベース表現型取得と時系列モデルを統合した収量予測手法を開発・検証しており、表現型取得ワークフローが中心的である。
abstractThis study introduces PheMuT, a novel phenology-informed, multi-modal time-series model that integrates visual and meteorological data streams to enhance strawberry yield forecasting.
MaizeLiDAR / point cloudLeafStem / branchMorphology / geometry measurementSegmentationLeaf traits
Maize leaf phenotypic parameters effectively reflect the photosynthesis and growth information of maize plants, which is crucial for breeding superior maize varieties. Current challenges include separating stems and leaves from a single maize plant and accurately measuring the phenotypic parameters of maize leaves. This study proposes a stem-leaf segmentation method based on region growing, incorporating adaptive cuboid region growing and slice region growing, alongside techniques for measuring phenotypic parameters of maize leaves. First, terrestrial laser scanning (TLS) was employed to obtain three-dimensional (3D) point cloud data of maize at the five-leaf (V5) and six-leaf (V6) stages. The point cloud data were then preprocessed to isolate single plant point clouds. Next, the maize point clouds were pre-segmented into three categories-central point clouds, partially expanded leaf point clouds, and unexpanded leaf point clouds-using center-edge segmentation, statistical filtering, and leaf classification. Adaptive cuboid region growing was applied to segment the unexpanded leaf point clouds, while slice region growing was used for partially expanded leaves, with Euclidean clustering optimizing the leaf point clouds, completing the segmentation process. Finally, various methods-including clustering counting, point-to-point distance accumulation, point-to-line distance, vector angle, point cloud triangulation, and triangle area accumulation-were utilized to automatically measure the number of maize leaves, leaf length, leaf width, leaf inclination angle, and leaf area. Compared with other point cloud stem-leaf segmentation methods based on geometric features and common 3D point cloud deep learning models (PointNet++, PointTransformer), the method proposed in this paper performs better. The segmentation results indicated that the Precision (P), Recall (R) and F₁-Score (F₁) for stem-leaf segmentation of all maize plants at the V5 stage exceeded 92.00%, with average values of 96.87%, 97.08%, and 96.97%, respectively. At the V6 stage, P, R, and F₁ exceeded 95.00%, with averages of 97.73%, 97.01%, and 97.67%, respectively. The algorithm accurately measured the number of leaves at the V5 stage, while a small error was noted at the V6 stage, yielding a percentage error (PE) of 0.93%. Measurement accuracy for leaf length, width, and area at both growth stages was greater than 93.80%, 92.80%, and 89.50%, respectively. Measurement accuracy for leaf inclination angle was lower, at 82.00% and 88.02% for the V5 and V6 stages, respectively. The proposed methods for stem-leaf segmentation and measurement of leaf phenotypic parameters are fast and accurate, providing technical support for high-quality breeding and intelligent management of maize. Our point cloud data of maize and source code is available from https://github.com/lmj-cau/stem-leaf-segmentation.git.
Why it matches plant phenotyping methodsトウモロコシの3D点群から茎葉を分割し、葉数・長さ・幅・面積・傾斜角を自動推定する手法を開発・比較検証しており、植物表現型取得が研究の中心である。
abstractThis study proposes a stem-leaf segmentation method based on region growing, incorporating adaptive cuboid region growing and slice region growing, alongside techniques for measuring phenotypic parameters of maize leaves.
TomatoLiDAR / point cloudLeafStem / branchMorphology / geometry measurementSegmentationLeaf traits
Tomatoes are a globally important horticultural crop, and their high-yield, high-quality breeding relies on high-throughput, precise phenotyping. While 3D point cloud technology offers a new avenue for non-destructive plant phenotyping, the inherent complexity of tomato plant organ morphology and growth dynamics poses a significant challenge to existing segmentation methods. To address this, this study employed multi-view RGB image reconstruction to cost-effectively acquire high-quality point cloud data from four growth cycles. Based on the characteristics of our data, we adapted and proposed a hybrid dual-path downsampling method (HDPD) for dataset augmentation, and constructed a dynamic reference point propagation network (DRP-Net) for semantic segmentation. The DRP-Net architecture addresses geometric feature mismatches between organs through a dynamic kernel edge convolution module (DKEC). Furthermore, it utilizes a global–local semantic feature fusion upsampling module (GL-SFFU) to overcome boundary blurring caused by plant growth and enhance detail discrimination. Based on the semantic segmentation results, a clustering algorithm was used to achieve leaf instance segmentation and extract key phenotypic parameters. Experimental results demonstrate that DRP-Net achieves significant performance in the tomato stem and leaf segmentation task, with mean precision, recall, F1 score, and mIoU reaching 94.97%, 93.93%, 94.43%, and 89.34%, respectively. The extracted phenotypic parameters, such as leaf length, leaf width, and leaf area, exhibit strong correlations with manual measurements (R² greater than 0.92 and 0.88, respectively). This study provides an effective technical solution for the precise segmentation of complex plant organs and high-throughput phenotyping analysis for breeding.
Why it matches plant phenotyping methodsトマトの3D点群から茎葉をセグメンテーションし、葉形質を抽出する手法を開発・検証しており、植物フェノタイピングが研究の中心である。
abstractconstructed a dynamic reference point propagation network (DRP-Net) for semantic segmentation
Accurate acquisition of phenotypic characteristics in protected crops is a crucial prerequisite for intelligent control and digital breeding in greenhouses. To accurately assess the phenotypic traits of protected lettuce, a specialized in situ phenotypic detection method has been developed. The Multimodal Features and Attention Mechanism for Phenotype Detection Model (MFAMNet) was developed for protected lettuce, employing a segmented multi-source image dataset for synchronous regression testing. The results revealed that the predicted values generated by MFAMNet exhibited a strong correlation with the measured values, achieving coefficients of determination of 0.96, 0.92, 0.95, 0.94, and 0.95 for plant height, crown width, leaf area, fresh weight, and dry weight, respectively. Ablation tests demonstrated that the deep learning detection framework based on multi-modal feature fusion significantly outperformed single-feature detection models, highlighting the advantages of integrating diverse data modalities. In addition, the multi-modal feature attention mechanism (MMF) facilitates both inter-modality and intra-modality interactions by capturing the global correlations between modalities and employing dynamic sparse spatial attention. The effectiveness of MMF has been validated through comparative experiments, demonstrating its suitability for the phenotypic detection of artificially cultivated lettuce. In summary, the method proposed in this study facilitates real-time monitoring of facility crops, enabling precise control of environmental parameters in protected agriculture and optimizing resource allocation. This approach contributes to the development of a comprehensive intelligent agriculture system and establishes a foundation for unmanned farms.
Why it matches plant phenotyping methodsレタスの草丈、株幅、葉面積、 fresh weight、dry weightを推定するマルチモーダル画像ベース手法を開発し、実測値との比較およびアブレーション・比較実験で検証しており、フェノタイピング手法が研究の中心である。
abstracta specialized in situ phenotypic detection method has been developed
Citrus is widely loved for its rich nutritional value and unique flavor, and also occupies an important place in agriculture and the economy. The citrus industry has suffered severe losses in recent years due to the proliferation of citrus Huanglongbing (HLB). The transmission of HLB occurs via the Asian citrus psyllid insect vector and grafting practices, with no efficacious therapeutic intervention identified to date, aside from mitigating its dissemination through prompt identification and eradication of infected citrus trees. Detection of HLB is difficult due to its incubation period and the variety of symptoms at different stages of infection. Hence, there exists a pressing requirement for a detection methodology capable of integrating multi-level features of HLB to facilitate precise identification of the disease across various stages of infection. Most of the existing assays use single sensing, which leads to limitations and incompleteness in identifying specific markers induced by HLB. In this study, the effective configuration and complementarity of multimodal sensory information is achieved by establishing a fusion and complementary mechanism at the level of pre-processing and analyzing multisource information. The extraction of computer vision and electronic nose features of citrus leaves was realized using custom-developed portable detection devices. The performance of HLB detection was compared on different datasets obtained by multimodal feature fusion methods which include direct fusion method, stepwise fusion method and the improved Recursive Feature Elimination and Cross Validation (RFECV) feature selection method. The improved RFECV feature selection method uses the RFECV algorithm for each classification step in the delineated stepwise classification model and performs the feature set preference by cross-validation. The final improved RFECV feature selection method worked best for fusion of visual and olfactory features with an accuracy of 95.38% for HLB samples at various symptomatic stages, with 94.23% for early stage HLB and 94.12% for Zn Def. & HLB-positive samples. Multimodal feature fusion for feature acquisition proved to be superior to feature acquisition from a single sensing source, with enhanced fusion of HLB-induced feature sets at the visual and olfactory levels. It helps to improve the stability of the HLB measurement model to achieve the detection of HLB samples in complex environments. This method can provide generalized technical support for the application of multi-source sensing information and multimodal feature fusion methods in plant disease detection.
Why it matches plant phenotyping methods柑橘葉の症状を対象に、カスタム開発した画像処理・電子鼻装置とマルチモーダル特徴融合によるHLB検出法を開発・評価しており、植物病態の取得・推定が研究の中心である。
abstractThe extraction of computer vision and electronic nose features of citrus leaves was realized using custom-developed portable detection devices.
Improving yield is one of the core goals of crop breeding. By predicting the potential yield of different breeding materials, breeders can screen these materials at different growth stages to select the best-performing breeding materials. However, existing yield prediction methods struggle to balance accuracy, interpretability, and robustness, often facing trade-offs between model complexity, data requirements, and generalizability. To address the above challenges, this study proposed a new hybrid method integrating remote sensing data assimilation and deep learning. The leaf area index was assimilated into the calibrated and validated WOFOST crop model using a newly designed data assimilation algorithm. The dataset, including partial outputs from the WOFOST model, development day, and vegetation indexes (VIs), was used to train the Temporal Fusion Transformer model for wheat yield prediction. The results showed that the new hybrid method achieved the highest performance in wheat yield prediction for different breeding materials in different study areas (R² of 0.831 and RMSE of 372.8 kg/ha in Yuhang experiment; R² of 0.704 and RMSE of 605.3 kg/ha in Zijingang experiment), which was better than other process-based model-driven methods and data-driven methods. This showed that the hybrid method had superior applicability in accurate yield prediction. Other results showed that increasing the number of data collections during the growth stage could significantly improve the performance of yield prediction and reduce error. Data from the middle and late growth stages contributed more to prediction performance than data from the early stages. Physiological variables and some VIs were the most important factors for yield prediction, while morphological characteristics contributed less. The importance and impact of each feature varied at different growth stages, highlighting the complex nonlinear relationship between characteristics and yield. In addition, since existing methods are difficult to fully utilize multi-source heterogeneous data related to yield in the breeding process, and the usability and user-friendliness of yield prediction software are also insufficient, an interactive yield prediction website has been developed based on a new hybrid method, a large language model (Llama), and related technologies to assist breeding decisions. This study aims to improve the efficiency of breeding material screening, provide an accurate, user-friendly, and well-interpretable yield prediction tool for wheat breeding, and facilitate smart breeding and decision making.
Why it matches plant phenotyping methodsコムギの収量という植物形質を、リモートセンシング・データ同化・深層学習で予測する手法を開発・比較検証し、育種向けソフトウェアも開発しているため、フェノタイピング手法が中心的である。
abstractThe results showed that the new hybrid method achieved the highest performance in wheat yield prediction for different breeding materials in different study areas
Efficient irrigation of horticultural crops under increasing water scarcity requires crop models that exploit high–resolution remote–sensing (RS) data. This study evaluated how unmanned aerial vehicle (UAV) multispectral and thermal observations improved AquaCrop-OSPy simulations of canopy cover (CC), actual evapotranspiration (ETₐ) and yield for irrigated broccoli under Mediterranean conditions. Broccoli was grown for two seasons in a 0.2 ha field in eastern Spain under two irrigation strategies: decision–support Irrigation Advisor (IA) versus farmer practice. A global sensitivity analysis (GSA) and two–stage calibration against Season 1 CC and yield identified canopy growth (CGC), harvest index (HIₒ) and transpiration phenology (GDDᵤₚ) as dominant controls; the calibrated model was validated in Season 2. UAV imagery provided CC via supervised classification and ETₐ via a two–source energy balance model (pyTSEB), which were assimilated into AquaCrop-OSPy on three dates using a hybrid observed–simulated scheme. Compared with lysimeter measurements, pyTSEB reproduced ETₐ with root–mean–square error (RMSE) 0.39 mm d⁻¹ and Nash–Sutcliffe efficiency (NSE) 0.93, whereas baseline AquaCrop-OSPy showed RMSE 1.24 mm d⁻¹ and NSE 0.27. Without assimilation, AquaCrop-OSPy reproduced mean yield but not subplot variability (RMSE 1.67 t ha⁻¹). Assimilating CC reduced yield RMSE by 8.9 %, ETₐ alone gave smaller gains, and joint CC + ETₐ assimilation achieved the lowest RMSE (1.47 t ha⁻¹, 11.9 % reduction). Across seasons, IA applied 20.6 % more water than farmer practice with no consistent yield or water–productivity benefits. These results, obtained within the limitations of this study, indicate that UAV-derived CC, complemented by ETₐ, modestly improves AquaCrop-OSPy yield predictions. Nevertheless, they should be interpreted as indicative rather than definitive and motivate further evaluations.
Why it matches plant phenotyping methodsUAV画像からブロッコリーの canopy cover と実蒸発散量を推定し、作物モデルへの同化性能を検証することが研究の中心であり、植物形質・状態の取得と技術評価に該当する。
abstractUAV imagery provided CC via supervised classification and ETₐ via a two–source energy balance model (pyTSEB), which were assimilated into AquaCrop-OSPy on three dates using a hybrid observed–simulated scheme.
Citrus anthracnose is a destructive fungal disease caused by Colletotrichum gloeosporioides, which causes leaf damage, fruit rot, and yield loss in citrus production. This study proposes an early detection method for citrus leaf anthracnose that integrates spectral and physiological data. Artificial inoculation experiments showed that the infected leaves exhibited yellowish-brown lesions, and the reflectance derived from visible-near-infrared (VNIR) spectroscopy and Fourier transform near-infrared (FTNIR) spectroscopy significantly decreased. Stomatal conductance and photosynthetic rate declined 4 days after inoculation. Physiological damage to leaves caused by fungal infection was more severe than mechanical damage. Three wavelength extraction algorithms [particle swarm optimization (PSO), bootstrapping soft shrinkage (BOSS), and least absolute shrinkage and selection operator (LASSO)] were combined with three machine learning models [artificial neural network (ANN), k-nearest neighbor (KNN), and categorical boosting (CatBoost)] to perform feature-level fusion on spectral data, photosynthetic parameters, and vegetation indices to improve classification accuracy. The fusion model had high classification accuracy (0.958–0.989) and Matthews correlation coefficient (MCC) (0.917–0.978). The model achieved the best performance in distinguishing leaves with early disease symptoms from healthy leaves, with an accuracy of 0.989, an F1 score of 0.989, and an MCC of 0.978. This research provides a reliable theoretical basis and technical support for the precise identification and early prevention and control of citrus anthracnose.
Why it matches plant phenotyping methods柑橘葉の病害状態をスペクトル・生理計測から推定する早期検出法を開発し、特徴抽出と機械学習モデルの性能を評価しており、植物表現型取得・推定が中心である。
abstractThis study proposes an early detection method for citrus leaf anthracnose that integrates spectral and physiological data.
Accurate estimation of the Leaf Area Index (LAI) is essential for assessing vegetation health and managing agricultural productivity. This study examines the application of Unmanned Aerial Vehicle (UAV)-based hyperspectral imaging and convolved EnMAP spectral data for estimating corn LAI, utilizing machine learning (ML) models to improve prediction accuracy. Various ML models, including k-nearest Neighbors (KNN), Support Vector Machines (SVM), Partial Least Squares Regression (PLS), and Random Forests (RF), were assessed to predict LAI from hyperspectral, EnMAP, and vegetation index features. Results demonstrate that PLS models consistently outperformed other ML approaches, achieving coefficients of determination (R²) ranging from 0.79 to 0.82. Notably, for the top two performing models (PLS and SVM) spectral indices such as NDRE, GNDVI, and NDVI proved more effective for LAI prediction than individual spectral bands. Interestingly, no matter the incorporation of hyperspectral wavelengths or EnMAP bands, the models predicting LAI were comparable. Feature importance analysis reinforced the dominance of vegetation indices as key predictors. The findings emphasize the benefits of high-resolution UAV hyperspectral imaging, convolved satellite spectral data, and machine learning, particularly PLS, for scalable and accurate LAI estimation in agroecosystems.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像と機械学習を用いてトウモロコシのLAIという植物形質を推定し、複数モデルと特徴量の性能を比較・評価しているため、形質取得手法が中心である。
abstractThis study examines the application of Unmanned Aerial Vehicle (UAV)-based hyperspectral imaging and convolved EnMAP spectral data for estimating corn LAI, utilizing machine learning (ML) models to improve prediction accuracy.
This study addresses the challenge of forecasting maize yield in southeastern Quebec by comparing weekly Unmanned Aerial Vehicle (UAV) and PlanetScope satellite imagery throughout the cropping season and across diverse growing conditions. Using five nitrogen treatments over three years with two sowing windows each year to generate variability within the dataset, eleven vegetation indices were evaluated to identify the best-performing indices and the optimal forecasting window. Indices were interpolated using curve fitting to enable evaluation at any stage of the growing season. Cross-validation simulated real-world application by excluding entire sowing events during model testing. Using a linear regression approach, results demonstrate that indices combining green and near-infrared bands (Green Normalized Difference Vegetation Index [GNDVI] and Chlorophyll Index Green [CIG]) exhibit superior forecasting potential compared to red-near-infrared (like NDVI) and RGB-based indices (like NGRDI). The optimal forecast window occurs during early grain filling (R2-R3 stages, around 2300 Crop Heat Units [CHU]), achieving Root Mean Square Coefficient of Variation (RMSCV) values of 12.51 % for UAVs and 15.28 % for PlanetScope. While PlanetScope maximum performance approached UAV capabilities, results showed a CHU range between 200 and 400 in the effective forecasting period (RMSCV < 20 %) compared to 850 to 1450 for UAV. For PlanetScope, adding multiple indices marginally improved precision, slightly reduced forecasting window and reduced model transferability. The analysis revealed weak correlations between indices and yield during early vegetative and senescence phases, indicating limited potential for enabling timely in-season management interventions. This study established UAV-based models as a reference point for assessing the limitations of satellite-derived forecasts.
Why it matches plant phenotyping methodsUAV・衛星画像と植生指数によるトウモロコシ収量推定を比較・交差検証しており、植物収量という形質の取得・予測手法が中心である。
abstractcomparing weekly Unmanned Aerial Vehicle (UAV) and PlanetScope satellite imagery throughout the cropping season
Accurate canopy photosynthesis modeling is essential for understanding and optimizing crop growth and yield in greenhouse agriculture. Current models have limited predictive capability due to inadequate responsiveness to dynamic environments and delays in parameter acquisition, making accurate predictions challenging under the complex conditions of solar greenhouses. This study aimed to develop a dynamic canopy photosynthesis model for greenhouse tomatoes, leveraging an IoT sensor network for real-time biological feedback and parameterization. By integrating real-time monitoring with dynamic feedback, the model facilitates precision management of greenhouse tomato cultivation, thereby optimizing plant growth, resource use efficiency, and yield predictability. To achieve this, a non-destructive inversion method based on a dual weighing system was developed, enabling accurate dynamic monitoring of tomato canopy leaf area index (LAI, R² ≥ 0.94) and the photosynthetic leaf area index (LAIₚ, R² ≥ 0.91), continuously providing parameters for updating modelling (validated against destructive sampling and actual measurements for trait specifics). Based on accurate parameter acquisition, a dynamic canopy photosynthesis model was developed using LAIₚ as the core variable, integrating above-canopy radiation. A newly developed parameter, which integrates the radiation component of transpiration, serves as a key factor for estimating photosynthesis. This innovative approach allows for accurate daily prediction and assessment of assimilated biomass. Experimental results from 2022 and 2023 showed that the LAIₚ model performed better than the comparison model, showing higher accuracy and adaptability (R² = 0.87 and 0.89, NRMSE = 0.17 and 0.12 vs. R² = 0.70 and 0.80, NRMSE = 0.26 and 0.15). These results confirmed the reliability of the integrated modeling framework, which forms a closed-loop system connecting real-time plant monitoring, statistical parameter inversion, online model adaptation, and biomass feedback verification. This modeling approach provides a solid foundation for precise growth simulation, sustainably improving yield and quality in solar greenhouse tomatoes, and advancing digital twin-enabled intelligent production.
Why it matches plant phenotyping methods植物キャノピーのLAIおよび光合成LAIを非破壊・連続推定するセンサー／逆解析法を開発し、破壊サンプリング等で検証している。植物形質取得とモデル連携が研究の中心である。
abstracta non-destructive inversion method based on a dual weighing system was developed, enabling accurate dynamic monitoring of tomato canopy leaf area index (LAI, R² ≥ 0.94) and the photosynthetic leaf area index (LAIₚ, R² ≥ 0.91)
The integration of electronic system into agricultural production can significantly enhance its efficiency and scalability. However, most of the current research focuses on the data acquisition and automated control. The development of expert-level, interpretable decision-making systems remains a challenge, primarily due to the prohibitive requirement for extensive domain-specific labeled data. In this manuscript, a novel agentic framework integrated with Large Language Models is proposed and demonstrated, using seedling assessment as a case study. The framework achieves high predictive accuracy, strong interpretability, and fast-adaption ability, offering a distinct advantage over methods that demand large labeled datasets. An agentic orchestration framework integrated with the Analytic Hierarchy Process and the reasoning ability of the Large Language Models is constructed to automatically derive the raw assessment rating. Based on a score calibration system using few-shot learning with three different types of lettuce, Butterhead, Grand Rapids, and Ramosa Hort, the final rating score can be derived with good prediction accuracy based on a small dataset (less than 20 labelled data). Additionally, three supplementary plant species (Sprout, Ball Brassica, and Rapa Brassica) are used to demonstrate the framework’s rapid adaptation capability. A field experiment guided by the agentic framework is conducted to prove that this seedling assessment system can be applied to help increase yield by more than 20 %. Our framework presents an important attempt towards an intelligent agricultural system that is capable to achieve expert-level and data-efficient decision making, thereby helping to bridge the critical gap between artificial intelligence research and practical agricultural application.
Why it matches plant phenotyping methodsLLMを用いた解釈可能なエージェント型フレームワークを開発し、苗の評価スコアという植物状態の推定・抽出に適用しているため、表現型取得・評価手法が研究の中心です。
titleFew-shot and interpretable agentic framework based on large language models for data-efficient plant phenotyping
This study proposes a generative AI-driven framework integrating CNN, VLM and LLM, aiming to provide intelligent solutions for the diagnosis of crop diseases and the generation of control strategies. The framework comprises four core modules: a data enhancement and style transfer module based on CycleGAN, a CNN-based disease detection module, a VLM-based visual semantic description module, and an LLM-based control strategy generation module. In the data enhancement module, CycleGAN is utilized to perform style transfer on the original dataset. This process makes the image features more realistic under natural conditions and improves the model's generalization ability in real-world agricultural production environments. In the disease detection module, the Yolo-CDDet model is developed, which adopts a cascaded feature learning architecture. This architecture consists of a deformable convolution backbone network, a global-local feature pyramid pooling neck network, and a decoupled prediction structure detection head, enabling precise identification and classification of disease regions. In the visual semantic description module, the MultiTask-CLIP model is constructed, featuring a multi-task classification head. The model outputs text descriptions with fixed feature combinations, providing detailed visual evidence for subsequent control strategy formulation. In the control strategy generation module, the Falcon-40B large language model serves as the core component. By leveraging web crawlers to collect open-source professional agricultural literature and applying the LoRA fine-tuning method to optimize model parameters, the model is optimized. It generates scientifically grounded and practical control recommendations tailored to specific disease characteristics. Experimental results demonstrate that the Yolo-CDDet model achieves superior performance on both the original dataset and the dataset enhanced by style transfer. The model exhibits high Recall, Average Precision, and other excellent metrics. The MultiTask-CLIP model outperforms competing models across multiple evaluation criteria, particularly excelling in CIDEr scores. Additionally, the control strategy generation mechanism based on Falcon-40B surpasses baseline models in terms of Recall, Precision, ROUGE-L, and other quantitative analysis indicators, producing high-quality control strategy texts. This study offers a novel and effective approach for the intelligent diagnosis and integrated management of crop diseases.
Why it matches plant phenotyping methods作物病害領域で、病斑領域の画像検出・分類モデルを開発し、データ拡張や複数モデルとの性能評価を行っているため、植物の病害状態を観測・推定する方法が中心である。
abstractThe framework comprises four core modules: a data enhancement and style transfer module based on CycleGAN, a CNN-based disease detection module, a VLM-based visual semantic description module, and an LLM-based control strategy generation module.
To address the high-throughput real-time detection requirements in industrial seed sorting scenarios, this study proposes an innovative solution coupling a lightweight detection algorithm with an industrial control system. By optimizing and integrating the YOLOv11-S architecture with the MobileNetV4 depth-wise separable convolution backbone, introducing the Focus operation for 4x downsampling via slicing concatenation without increasing computation, and embedding a mixed local channel attention mechanism, an industrially applicable model, Anomalous Seed Detection-YOLO(ASD-YOLO), with a parameter size of only 9.5 MB, was constructed. This model achieves a mean average precision (mAP) of 96.5 % while reaching a maximum processing capability of 62 FPS on a single device. Simultaneously, by incorporating algorithms such as a feedback error correction mechanism developed in conjunction with an industrial-grade pulse coordination control mechanism, the system achieves stable end-to-end latency control at the 35 ms level in a pepper seed anomaly detection production line environment. It supports continuous 24-h stable operation at a throughput of 10,000 seeds/min, with a relative error controlled to 3.3 mm. Based on the detection results, a fuzzy grading algorithm was developed to categorize the seed quality into five levels using membership functions. This provides a quantitative basis for refined storage management and differentiated processing, achieving a statistically significant 16.2 % reduction in the misjudgment rate compared with traditional grading methods. By constructing an “artificial intelligent algorithm-pulse coordination-protocol coupling” trinity architecture, the proposed model establishes a universal methodological framework for lightweight model deployment in agricultural intelligent manufacturing scenarios, offering a scalable standardized solution for seed quality control.
Why it matches plant phenotyping methods種子の異常を画像検出し品質を5段階評価する軽量モデルと産業用制御システムを開発しており、植物器官の状態取得・抽出が研究の中心である。
abstractthis study proposes an innovative solution coupling a lightweight detection algorithm with an industrial control system
Accurate and temporally consistent multispectral observations are essential for monitoring alfalfa yield and quality, given its frequent harvest cycles and rapid regrowth. However, optical satellite imagery is often constrained by cloud cover, revisit intervals, and sensor availability. To overcome these limitations, we propose a novel Alfalfa Multimodal Generative Adversarial Network (AMGAN) designed for near-daily multispectral image reconstruction. Unlike conventional image-to-image or spatiotemporal fusion methods that overlook crop-specific characteristics, are restricted to observed timestamps, or depend heavily on dense temporal series, AMGAN leverages multisource (Landsat-8/9, Sentinel-1, PlanetScope) and multimodal (climate, geographic, temporal) information within an adversarial learning paradigm. This enables high-quality image generation from minimal inputs. Extensive experiments across five major alfalfa-producing states in the United States (2022-2024) show that AMGAN consistently surpasses four state-of-the-art (SOTA) deep learning baselines. It achieves higher reconstruction accuracy across all spectral bands, with pronounced gains in red-edge and near-infrared (NIR) regions critical for vegetation assessment. Multisource integration and multimodal cues enhance robustness, ensuring reliable performance under diverse observation scenarios. The reconstructed imagery was subsequently evaluated in alfalfa yield and quality prediction tasks. Results demonstrated high predictive accuracy for dry matter yield (DM) in the cross validation (CV) experiment with a coefficient of determination (R²) of 0.80, and moderate correlations for selected quality traits such as crude protein (CP), non-fiber carbohydrates (NFC), and minerals, while nutritive value traits tied to complex biochemical processes remained more challenging. Overall, this study underscores the potential of multimodal adversarial learning to bridge observational gaps in alfalfa monitoring. The proposed framework provides a scalable, crop-specific approach for generating temporally dense imagery, supporting precision management for biomass-related and proximate quality traits, while performance for digestibility traits remains limited.
Why it matches plant phenotyping methodsアルファルファの収量・品質形質推定を支えるマルチスペクトル画像再構成手法を開発し、複数手法との比較検証と形質予測評価を行っており、フェノタイピング手法が中心である。
abstractwe propose a novel Alfalfa Multimodal Generative Adversarial Network (AMGAN) designed for near-daily multispectral image reconstruction.
Efficient management and precise monitoring are essential for the sustainable control of crop diseases and pests. Traditional unimodal methods exhibit reduced reliability due to data gaps and environmental fluctuations. Multimodal artificial intelligence (AI) offers a promising alternative by integrating complementary data sources and enhancing robustness and adaptability. However, a comprehensive synthesis connecting multimodal AI with multi-scale disease and pest management is still lacking. Based on 950 publications from the past decade reflecting a 31.7% annual growth rate over the past five years, this review examines the evolution of AI-driven research and compares unimodal and multimodal approaches by summarizing major data modalities, fusion strategies, and modeling techniques. Deep learning emerges as the most widely used class of AI methods, and quantitative evidence indicates that multimodal systems achieve approximately 3–48.9% higher diagnostic accuracy than unimodal models. Evidence from 27 studies demonstrates the effectiveness of multimodal fusion across imaging, spectral, environmental, and sensor-based datasets. Building upon these findings, we propose a novel three-level management framework comprising point-level diagnosis, area-scale monitoring, and spatiotemporal forecasting, clarifying how multimodal AI strengthens each task. We further highlight the role of Plant Electronic Medical Records (PEMRs) and outline a conceptual virtual plant clinic to support continuous, data-driven crop health services. Finally, this review identifies key directions including advanced fusion strategies, lightweight and interpretable models, digital twin integration, and scalable decision-support systems, which are essential for intelligent and sustainable crop disease and pest management.
Why it matches plant phenotyping methods植物病害の診断・モニタリングを対象に、画像・スペクトル・環境・センサーデータを統合するAI手法とその性能を体系的にレビューしており、植物の病害状態推定が中心的な方法論的貢献である。
abstractthis review examines the evolution of AI-driven research and compares unimodal and multimodal approaches by summarizing major data modalities, fusion strategies, and modeling techniques.
Sustainable plant cultivation is critical for supporting long-duration space missions by ensuring reliable food production in extraterrestrial environments where resources are severely limited and growth systems operate in closed-loop conditions. With crew members managing multiple critical mission tasks and having minimal time for plant care, autonomous stress detection systems must provide reliable, interpretable diagnostics to enable rapid, informed decision-making for crop management. This study utilized a custom hyperspectral imaging (HSI) system designed for space applications to develop an AI-driven diagnostic framework. We propose a novel SAM-ViT-3PE architecture that uniquely combines sparse spectral band selection with 3D spatial-spectral patch embedding, preserving rich spatial-spectral information typically lost in conventional ROI-averaged approaches. A key temporal finding identified Day 3 After Treatment (DAT 3) as the critical threshold where drought stress signatures become distinctly detectable, with accuracy dramatically improving from 72.2% to 95.9%. By focusing analysis on data from DAT 3 onward, the SAM-ViT-3PE model achieved superior performance compared to traditional ML methods and standard deep learning approaches, with accuracy of 95.4%, precision of 96.6% and recall of 94.1%. Furthermore, Explainable AI using Integrated Gradients enabled interpretable diagnostics through physiologically meaningful spectral bands and spatial stress patterns. These results demonstrate that the AI-enhanced HSI framework provides both high-accuracy autonomous detection and scientifically grounded interpretability essential for trustworthy crop management in resource-constrained space environments.
Why it matches plant phenotyping methods植物の干ばつストレス状態を hyperspectral imaging と深層学習で検出する診断フレームワークを開発・評価しており、植物表現型取得が研究の中心である。
abstractThis study utilized a custom hyperspectral imaging (HSI) system designed for space applications to develop an AI-driven diagnostic framework.
Timely and precise harvest scheduling is critical for maintaining tea quality and improving labor efficiency. This study aimed to develop an integrated Internet of Things (IoT) and artificial intelligence (AI) framework for automated monitoring and growth modeling of tea shoots, enabling data-driven plantation management. Solar-powered Plantation Monitoring Systems (PMS) were deployed to continuously capture canopy images and environmental data, reducing reliance on manual inspections. An enhanced YOLOv11 segmentation model, incorporating HSI color space conversion, monocular depth estimation, and shape-based temporal tracking, was used to detect pluckable tea shoots with high accuracy. The computed Tea Shoot Density Index (TSDI) showed strong agreement with ground truth measurements (RMSE = 2.542, R² = 0.931). Three sigmoid growth models − 3PL, 4PL, and Gompertz − were evaluated using growing degree days (GDD) as the time scale. The 4PL model achieved the best performance (RMSE = 0.698, R² = 0.897) and predicted optimal harvest timing with a mean absolute error (MAE) of 2.7 days, while offering interpretable parameters that reflect shoot retention, growth rate, and maturation dynamics. These parameters provided actionable insights for optimizing irrigation, fertilization, and harvest scheduling across different growth stages. The proposed system delivers a scalable and automated solution for precision tea agriculture, enhancing productivity, improving tea quality, and supporting the transition from experience-based to data-driven management.
Why it matches plant phenotyping methods茶園画像から摘採可能な新芽を検出し、密度指標と生育・成熟状態を推定するIoT・AI計測手法の開発と精度検証が研究の中心であるため。
abstractThis study aimed to develop an integrated Internet of Things (IoT) and artificial intelligence (AI) framework for automated monitoring and growth modeling of tea shoots
Thermal imaging is becoming a valuable tool for monitoring plant canopy temperature, which can serve as an indicator of crop water stress. However, specialized thermal sensors are often cost-prohibitive. This study explored strategies for supplementing crop water stress monitoring by generating synthetic thermal images from standard Red-Green-Blue (RGB) imagery captured using an unmanned aerial vehicle system (UAVs) equipped with a Zenmuse XT2 sensor and leveraging deep learning models. UAV-based RGB and thermal images were collected from 32 experimental plots of sweet corn and green beans over three growing seasons from 2020 to 2023. Each crop was subjected to one full and three deficit irrigation treatments, replicated four times. A total of 3,400 UAV images were collected over three seasons. Image processing was done in Pix4D software, and orthomosaic RGB and thermal maps were spatially aligned using ground control points (GCPs). The UAV RGB and thermal map data were split into 80 % and 20 % for training and testing, respectively. Two image-to-image translation generative adversarial network (GAN) deep learning models, specifically Pix2PixGAN and CycleGAN, were used to generate synthetic thermal images from RGB inputs. Image quality evaluation metrics, i.e., correlation coefficients (r), mean squared error (MSE), peak signal-to-noise ratio (PSNR), and structural similarity index (SSIM), were used to evaluate the models’ performance. Crop water stress index (CWSI) values were also computed from measured and generated thermal imageries to assess practical applicability. Generated thermal canopy temperature outputs from the Pix2PixGAN model showed a strong correlation with the measured data using a thermal camera (r >0.95). Moreover, Pix2PixGAN resulted in lower MSE (5.63) and higher PSNR (42.98) than CycleGAN (MSE = 7.09, PSNR = 40.56), whereas CycleGAN had a slightly higher SSIM (0.44) than Pix2PixGAN (0.31). CWSI values derived from the generated thermal images reflected the expected gradients of water stress across irrigation treatments, with the highest CSWI observed from deficit irrigation treatments compared to the full irrigation. These results demonstrate that RGB-to-synthetic-thermal image translation using GAN models could be used to support crop water stress assessment and irrigation scheduling.
Why it matches plant phenotyping methodsRGB画像から合成熱画像を生成し、作物キャノピー温度と水ストレス指標を推定するGAN手法の開発・比較・性能評価が中心であり、植物状態の表現型取得に直接関係する。
abstractTwo image-to-image translation generative adversarial network (GAN) deep learning models, specifically Pix2PixGAN and CycleGAN, were used to generate synthetic thermal images from RGB inputs.
Accurate growth stage recognition is vital for optimising crop inputs and improving yield efficiency. However, single-modal methods often fail to distinguish phenologically adjacent stages due to canopy similarity, spectral saturation, or structural ambiguity, leading to mistimed agronomic actions and resource loss. Moreover, the high computational demands of complex deep neural networks hinder practical implementation. To address this challenge, this paper proposes a lightweight multimodal data fusion network for wheat growth stage recognition. Specifically, the RGB images, multispectral (MS) data, and digital surface model (DSM) acquired by Unmanned Aerial Vehicles (UAV), along with derived spectral vegetation index (VI), are used to capture multimodal canopy features, including colour, spectral reflectance, and spatial structure. Furthermore, the approach utilises MobileNetV3-Small, a lightweight convolutional neural network, as the backbone to construct the multimodal data fusion framework. This framework enables efficient feature extraction and integration, achieving precise and rapid wheat growth stage recognition with minimal computational overhead. The results demonstrate that, compared to single-modal models, the proposed multimodal fusion model significantly enhances growth stage recognition accuracy, achieving an accuracy of 99.57 %, a precision of 99.58 %, a recall of 99.57 %, and an F1 score of 99.57 %. Notably, it improves stage differentiation in critical transitions such as Booting to Heading, reducing field misclassification risks and supporting quick decision-making. Comparative analysis with MobileNetV3-Large, ResNet-18, MNASNet, EfficientNet-B0, and ConvNeXt-Tiny demonstrates that MobileNetV3-Small offers the best trade-off between accuracy and resource efficiency, with only 1.53 M parameters and an inference time of 6.03 ms on RTX 4090 and 25.11 ms on Jetson Orin NX. This efficiency enables real-time deployment on resource-constrained edge devices, such as onboard UAV processors or in-field embedded systems. Overall, this study effectively overcomes the challenges of recognising adjacent growth stages and computational constraints, offering a robust theoretical foundation and an efficient, accurate solution for wheat growth stage recognition.
Why it matches plant phenotyping methodsUAV画像・マルチスペクトル・DSMを用いて小麦の生育段階という植物状態を推定する融合手法を開発・評価しており、表現型取得・抽出が研究の中心である。
abstractthis paper proposes a lightweight multimodal data fusion network for wheat growth stage recognition.
Phosphorus (P) is a vital macronutrient necessary for synthesizing essential plant biomolecules. Accurate identification of plant P deficiency symptoms is critical for effective crop management and optimizing crop yield. Hyperspectral sensing provides a real-time, non-destructive avenue for assessing crop nutrient status, while its performance largely depends on the representativeness of the extracted features. In this study, a handheld proximal transmittance hyperspectral imager, LeafSpec, was utilized to collect leaf-level hyperspectral images at corn V6 vegetative stage. A novel feature mining algorithm was proposed to extract and combine the spatial and spectral features in visible and near-infrared range, enabling effective differentiation of P deficiency. The correlation coefficient between the P content and the selected spatial-spectral features reached 0.77. Compared with spectral indices, the combined spatial-spectral feature showed a more significant differences among corn plants under different P treatments, especially between the medium and sufficient P levels. Feature visualization heatmaps, highlighting leaf venation variations with spatial-spectral calculations, provided direct evidences of effectiveness. This study shows the potential of integrating handheld proximal transmittance hyperspectral imaging with feature mining algorithm for early-stage differentiation of P levels in corn plants.
Why it matches plant phenotyping methodsトウモロコシ葉のリン欠乏状態を対象に、ハイパースペクトル画像取得と空間・スペクトル特徴マイニング手法を開発・評価しており、表現型状態の推定法が研究の中心である。
abstractA novel feature mining algorithm was proposed to extract and combine the spatial and spectral features in visible and near-infrared range, enabling effective differentiation of P deficiency.
Accurate real-time quantification of strawberry ripeness is critical for advancing selective strawberry harvesting robots. However, existing methods often overlook inconsistencies caused by color distortion under varying light intensities. This study aims to verify and quantitatively analyze the influence of illumination on strawberry ripeness, and further presents a novel light-resilient vision-based ripeness regression method that overcomes these limitations through three key innovations. First, the research established the first fine-grained ripeness metric under controlled lighting conditions and introduced the comprehensive LightStrawberry dataset, featuring multi-illumination strawberry images. Second, we propose SRR-Net, an innovative end-to-end strawberry ripeness regression network built upon YOLOv8/YOLOv11. The network incorporates a dedicated ripeness regression branch that operates in parallel with the detection and segmentation heads, enabling simultaneous and efficient estimation of strawberry maturity. To further mitigate lighting-induced color distortion, RetinexNet was integrated to decompose, adjust, and reconstruct images by normalizing illumination and reflectance. Experiments demonstrated that SRR-Net achieved 0.918 mAP@50 for segmentation and operated at 210.3 FPS based on YOLOv11, while SRR-Net with RetinexNet attained 0.898 mAP@50 and 43.39 FPS. Though slightly lower in precision than other methods, both significantly improved ripeness accuracy, with mean absolute errors (MAE) of 0.040 and 0.037, representing 68.75 % and 71.09 % improvements over conventional Mask R-CNN approaches. Field experiments further demonstrated that SRR-Net and SRR-Net with RetinexNet achieved superior performance in ripeness regression. However, their detection and segmentation performance showed limited adaptability to real orchard conditions due to the characteristics of the LightStrawberry dataset. Overall, both models outperformed the standard YOLOv8/v11 baselines but were slightly inferior to Mask R-CNN. This work provides a robust solution for strawberry-harvesting robotics, enabling reliable ripeness assessment in challenging field environments.
Why it matches plant phenotyping methods画像からイチゴの成熟度を定量推定する回帰手法、照明補正、データセットを開発・評価しており、植物形質取得が中心である。
abstractpresents a novel light-resilient vision-based ripeness regression method
Deep learning has achieved promising performance for fruit and vegetable image analysis, by possessing strong representation power, and providing resilient generalization and broad transferability on large-scale data for classification, detection, and segmentation tasks, which is indispensable role in optimizing agricultural practices. This comprehensive survey reviews over 270 recent studies, offering a deep exploration of the key techniques and strategies, fundamental properties, and advancements and future directions according to different categories of deep learning methods for fruit and vegetable image analysis. Furthermore, this paper outlines the novelty and concept of fruit and vegetable image analysis, summarizes publicly available datasets, evaluation metrics, and discusses successful applications in disease detection, quality grading, yield estimation, localization, and multiple application integration. The survey emphasizes the need for processing large-scale datasets and exploring the potential of efficient deep learning for enhancing real-time applications and specific tasks. By comprehensively comparing and analyzing the fundamental attributes of the fruit and vegetable image analysis methods from a fresh perspective, this survey reveals the commonalities and disparities of divert techniques and guides researchers and practitioners toward developing more efficient and accurate solutions.
Why it matches plant phenotyping methods果実・野菜画像解析の深層学習手法を包括的にレビューし、疾患検出や収量推定など植物の状態・形質推定、データセット、評価指標を扱うため、フェノタイピング手法レビューとして中心的です。
abstractThis comprehensive survey reviews over 270 recent studies
Close-range spectral imaging provides technical support for leaf detection during the early infection process of SCLB (southern corn leaf blight). However, due to the randomness of pathogen infection and the low visibility of early lesions, the temporal spectral signals obtained using this technique have poor continuity and low sensitivity. To improve the ability of temporal spectral signals to detect early-stage infection, this study proposes a signal extraction method based on reverse temporal spectral image matching, and a signal decoupling method based on DSO-CWT (decomposition of scale-optimized continuous wavelet transform) is also proposed to improve signal sensitivity. First, the spectral images were preprocessed. Image entropy was used to quantify the changing patterns of symptoms in early SCLB infection. Second, a temporal spectral signal extraction method based on ASpanFormer (adaptive span transformer) reverse temporal spectral image matching is proposed. The calculation results of LPIPS (learned perceptual image patch similarity) indicates that the average matching error between adjacent periods is less than 0.2, which suggests that this method can enhance the extraction accuracy of weak temporal spectral signals in early infection. Third, the T-test was used to evaluate the detection sensitivity of temporal spectral signals at different early stages of infection. The results showed that the temporal spectral signal still had low sensitivity for detecting different stages of infection. Therefore, a temporal signal decoupling method based on DSO-CWT is proposed, which enhances the detection sensitivity of temporal spectral signals by performing time–frequency domain conversion. Finally, a diagnostic model for the early SCLB infection was established by fusing fluorescence and reflectance spectral signals. After DSO-CWT processing, the accuracy of the modelling set improved from 47.62% to 94.22%, and the accuracy of the validation set was improved from 47.62% to 91.27%. This study improves the extraction accuracy and detection sensitivity of temporal spectral signals for early SCLB infection by using signal extraction based on reverse temporal spectral image matching and deep signal decoupling based on DSO-CWT, providing new insights for early detection of SCLB infection.
Why it matches plant phenotyping methods植物病害の感染状態を対象に、時系列スペクトル信号の抽出・分離手法を開発し、検出感度と診断精度を検証しているため、病徴状態のフェノタイピング手法が中心です。
abstractthis study proposes a signal extraction method based on reverse temporal spectral image matching, and a signal decoupling method based on DSO-CWT
Hyperspectral imaging (HSI) has emerged as a powerful tool for precision agriculture, enabling the non-destructive monitoring of crop biochemical and physiological traits. However, HSI alone lacks structural context, which limits its ability to accurately capture complex canopy architectures and organ-level traits. Integrating HSI with depth-sensing modalities such as Light Detection and Ranging (LiDAR), Red, Green, Blue, and Depth (RGB-D) cameras, and computational reconstruction technique such as photogrammetry enables the generation of three-dimensional hyperspectral point clouds, combining spectral richness with geometric fidelity. This multi-modal fusion enhances crop trait estimation, including biomass, leaf chlorophyll content, canopy height, leaf area, and stress indicators, while improving the robustness of phenotyping under occlusions, shadows, and varying illumination. Dimensionality reduction, feature selection, and machine learning approaches, including deep learning and explainable AI, are useful for handling high-dimensional hyperspectral data and extracting actionable agronomic insights. Moreover, the integration of thermal, radar, and Global Navigation Satellite System (GNSS) data further expands the capabilities of multi-modal sensing, enabling continuous, all-weather crop monitoring and accurate spatial referencing. Despite these advances, most studies to date focus on controlled environments, highlighting the need for field-based validation to ensure the reliability and scalability of HSI-depth fusion techniques. This review consolidates current knowledge on multi-modal hyperspectral and 3D crop reconstruction, highlighting methods, applications, and challenges, and outlines future directions for implementing high-throughput, real-time phenotyping and precision agriculture solutions.
Why it matches plant phenotyping methods植物形質推定のためのハイパースペクトル・深度センシング・3D再構成手法を中心に扱うレビューであり、フェノタイピング手法レビューに該当する。
abstractThis review consolidates current knowledge on multi-modal hyperspectral and 3D crop reconstruction, highlighting methods, applications, and challenges
Vertical farming offers a promising solution to global food security and urbanization challenges, yet its widespread adoption is hindered by high costs, particularly for lighting. Addressing this requires enhancing light use efficiency (LUE) through intelligent control strategies. While numerous studies have investigated the effects of light intensity on lettuce growth, relatively few have explored the potential benefits of stage-specific light regulation. In this study, we first developed an automated 3D phenotyping pipeline based on multi-view reconstruction to quantify canopy morphology and light interception. Utilizing this quantitative framework, we conducted a dynamic light experiment with lettuce in a commercial plant factory to evaluate four dynamic light-intensity strategies. The proposed 3D phenotyping pipeline demonstrated promising performance for canopy information extraction, with RMSEs for plant height, canopy diameter, and projected leaf area of 0.79 cm, 1.05 cm, and 44.3 cm², respectively. The “high-low-high” dynamic lighting strategy, applying higher light intensity during the early and late growth stages and lower intensity during the mid-growth stage, successfully optimized canopy morphology for better light capture. This treatment significantly increased shoot fresh and dry weights by 28 % and 65 %, respectively, compared to constant lighting. Furthermore, it enhanced LUE based on incident and intercepted light integrals by 67 % and 19 %, while reducing electricity consumption per unit of fresh weight by 24 %. Nutritional quality analysis showed the treatment increased soluble sugars and starch contents. By integrating advanced 3D phenotyping with dynamic light intensity control, this study demonstrates a prototype for intelligent decision-making to enhance yield and energy use efficiency in practical vertical farming.
Why it matches plant phenotyping methods自動3Dフェノタイピングパイプラインを開発し、マルチビュー再構成で植物体形態と光遮断を定量化、精度評価も実施しており、フェノタイピング手法が研究の中心である。
abstractwe first developed an automated 3D phenotyping pipeline based on multi-view reconstruction to quantify canopy morphology and light interception.
The center-pivot irrigation system is a highly efficient water-saving agricultural system. However, it typically operates solely based on predefined paths, speeds, and water volumes, and cannot assess the true needs of the crops due to its inability to monitor their growth status. To address this limitation, we deployed cameras on the center-pivot irrigation system to establish a mobile phenotyping platform, and developed a maize tassel detection model based on the YOLOv11 architecture. Several key improvements were implemented to address the complex morphology and challenging feature extraction of maize tassels: The ODConv (Omni-Dimensional Convolution) module was introduced to more comprehensively capture dynamic tassel features through a dynamic convolution strategy; To suppress interference from complex field environments on detection results, the SE (Squeeze-and-Excitation) attention mechanism was embedded to enhance effective feature responses and reduce noise impact; The BiFPN (Weighted Bi-directional Feature Pyramid Network) structure was adopted to strengthen multi-scale feature fusion capabilities, further improving the model’s detection performance for tassels at various scales; To tackle the difficulty of tassel recognition and localization in field environments, the C3K2 module was fused with the CSAM (Cross-Slice Attention Mechanism) to construct a C3K2-CSAM module, achieving more precise tassel identification and localization. Experimental results demonstrate that the OSBC-YOLO model achieved a precision of 91.8 % and a mean average precision (mAP) of 85.7 %, while reducing the number of parameters, FLOPs, and memory usage by 8.1 %, 28.1 %, and 7.3 % respectively compared to the original model. Field tests conducted on the center-pivot irrigation system revealed an error rate of 5.87 % between the number of tassels detected by the model and the ground truth count. These results verify that the OSBC-YOLO model can effectively perform tassel counting in practical operating environments and possesses the capability for recognition of crop phenotypic traits. This system provides critical data support for intelligent variable-rate irrigation decisions, promoting the transformation of center-pivot irrigation systems from traditional “single-function operation” to an integrated “perception-decision-execution” mode.
Why it matches plant phenotyping methodsトウモロコシ雄穂の検出・計数を行う移動型画像表現型解析プラットフォームとモデルを開発・検証しており、植物形態形質の取得が中心的貢献である。
abstractwe deployed cameras on the center-pivot irrigation system to establish a mobile phenotyping platform, and developed a maize tassel detection model based on the YOLOv11 architecture
This study investigates the application of hyperspectral imaging and machine learning techniques for detecting and classifying salinity stress in canola (Brassica napus L.). After analysis of various methods, we employed a ridge classifier model to categorize six classes of salinity, utilizing various spectral bands and vegetation indices across multiple model iterations. Spectral signature analysis revealed significant changes in reflectance patterns for wavelengths exceeding 740 nm, corresponding to the near-infrared (NIR) region. We developed two novel vegetation indices tailored for salinity stress detection, which, when combined with established indices and selected spectral bands, significantly improved classification accuracy. Our sequential model refinement process demonstrated incremental improvements in accuracy, with the final model achieving 82.61 % accuracy on the test set using only 15 features. This represents a substantial reduction from the initial 331 features while maintaining high accuracy. The most effective features primarily spanned wavelengths corresponding to Sentinel-2A bands, with notable exceptions at 405.04 nm and 983.96 nm. Comparison with Sentinel-2 spectral bands revealed that while some important wavelengths align with the satellite sensor’s capabilities, several fall outside its capture range. Notably, our findings suggest that Sentinel-2 bands B1, B5, B6, B7, and B9 may have limited efficacy in identifying salinity stress in canola, highlighting the potential for crop-specific optimization of spectral bands in remote sensing applications. This comprehensive analysis provides insights into the most effective spectral regions and vegetation indices for salinity classification in canola, offering the potential for improved precision agriculture practices. Our findings contribute to the growing body of knowledge on non-invasive crop stress detection and pave the way for future research in hyperspectral imaging applications for sustainable agriculture.
Why it matches plant phenotyping methodsハイパースペクトル画像と機械学習を用いて、カノーラの塩ストレスを検出・分類する手法を開発・評価しており、植物状態の取得・推定が研究の中心である。
abstractThis study investigates the application of hyperspectral imaging and machine learning techniques for detecting and classifying salinity stress in canola (Brassica napus L.).
LettuceTissuePhysiological trait estimationWater status / transpiration
With the rapid development of smart agriculture, the agricultural Internet of Things (Ag-IoT) has gradually established a monitoring system centered on distributed sensing, low-power communication, and intelligent control. However, current solutions underexplore the perceptible characteristics of communication signals, and therefore do not fully utilize their latent potential in environmental perception. This paper targets the demand for crop water monitoring and introduces an integrated sensing and communication (ISAC) approach. This method can achieve non-contact and continuous perception of crop water status by reusing the communication link without altering the existing hardware architecture and frequency band configuration. Taking leafy vegetables such as lettuce as the research object, a prototype system based on a 3 GHz communication link was built. A quantitative mapping model between the water content of plant tissues and the amplitude and phase disturbances they cause to electromagnetic waves was established. A joint optimization mechanism that considers both communication performance and sensing accuracy was proposed to achieve a coordinated configuration between communication quality (BER < 10⁻⁴, SNR ≈ 20 dB, EVM < 8 %) and sensing accuracy (MAE = 2.51 %, R² = 0.92). Experiments were conducted in controlled environments and production-like scenarios, demonstrating that the method can stably identify the water status of lettuce while ensuring communication quality of service (QoS). The proposed ISAC method is potentially compatible with existing Ag-IoT frequency bands and physical-layer infrastructures, assuming access to pilot/CSI and airtime control. Protocol-level integration with LoRa, Wi-Fi, and NB-IoT is defined as future work It provides a low-cost, high-integration, and easily scalable communication-driven solution for water monitoring in smart agriculture.
Why it matches plant phenotyping methodsレタスの水分状態という植物生理形質を、通信信号を再利用した非接触センシングで推定する方法を開発し、プロトタイプと精度検証を行っており、フェノタイピング手法が中心である。
abstractThis method can achieve non-contact and continuous perception of crop water status by reusing the communication link
Sweetpotato is a crucial food crop globally, valued for both its economic significance and health benefits. However, the prevalence of sweetpotato virus diseases (SPVD) poses a serious threat to the industry, leading to reduced yields and economic losses for farmers. Efficient diagnostic techniques are essential for ensuring food security and consumer health. Traditional diagnostic methods are effective but suffer from complexity, time consumption, and high costs. To address these challenges, a novel real-time end-to-end detector called SPVD-DETR based on the Transformer architecture is proposed in this study. By utilizing the unmanned aerial vehicle (UAV) orthomosaic image, SPVD can be diagnosed in real-time at the field scale. First, aerial survey tasks are customized with automated drone tools to rapidly scan sweetpotato fields, generating high-resolution orthophotos and a stitched orthomosaic image for analysis. Then, the object detector is enhanced by incorporating efficient backbones and hybrid encoder modules such as cascaded group self-attention, attention-based scale fusion, and dynamic upsampling. Extensive ablation studies and comparative results show that SPVD-DETR achieves a good balance between real-time performance and accuracy. Next, the model is fine-tuned on the SPVD image tiles and achieves a detection accuracy of 31.3% mean average precision (mAP) with the fastest inference speed of 90 frames per second (FPS). Finally, the prediction results are mapped back to the orthomosaic image, estimating an overall SPVD incidence rate of 15% with a misdiagnosis rate of 14%. This study introduces a novel paradigm for detecting SPVD at the field scale, promoting automatic and intelligent plant disease detection for large-scale high-throughput phenotyping in precision agriculture.
Why it matches plant phenotyping methodsUAV画像からサツマイモウイルス病の発生・罹病状態を推定する検出手法を開発し、アブレーション、比較評価、精度・速度・誤診率を検証しており、植物表現型取得が中心である。
abstracta novel real-time end-to-end detector called SPVD-DETR based on the Transformer architecture is proposed in this study.
WheatField / plotRGB / grayscaleThermalPanicle / ear / spikeSeed / grainPhysiological trait estimationSegmentationGrowth / development / phenologyWater status / transpiration
Grain filling plays a vital role in determining both the yield and quality of wheat. Therefore, timely and accurate monitoring of the grain filling course (GFC) is essential for assessing the feasibility of harvest timing optimization. Traditional methods based on field sampling are time-consuming and destructive. This study presents a non-destructive method for estimating the wheat grain filling course (GFC) by integrating ground-based RGB and thermal infrared imagery. Wheat ears were first segmented using a temperature-threshold approach, after which colour and temperature features were extracted. Grain water content (GWC) was then estimated using a Normalised Relative Ear Temperature (NRET) index, while days after anthesis (DAA) were retrieved using a piecewise linear model derived from ear colour features. Finally, a grain filling index (Kf) was developed using DAA corresponding to 25 % moisture content (DAA25%) to quantify the GFC. Results showed that thermal images acquired at 17:00 showed the greatest separability between ears and background canopy and the highest sensitivity to irrigation differences. Both NRET and DAA based models provided accurate GWC estimates (R² = 0.86 and 0.91; RMSE = 3.13 % and 4.21 %; rRMSE = 0.07 and 0.09, respectively). The Kf index effectively captured differences in GFC under different irrigation treatments and detected early maturity under water stress (p < 0.05). This study demonstrates the potential of combining thermal and RGB imagery for high-resolution, non-destructive monitoring of wheat grain filling and for supporting timely harvest management.
Why it matches plant phenotyping methodsRGB画像と熱赤外画像を統合し、穂の分割・特徴抽出から穀粒水分含量と登熟進行を推定する非破壊フェノタイピング手法が研究の中心である。
abstractThis study presents a non-destructive method for estimating the wheat grain filling course (GFC) by integrating ground-based RGB and thermal infrared imagery.
Plant motion provides valuable indicators of physiological responses to water stress. In this study, we present a structured image-based approach to define and test morpho-kinematic (MK) traits from lettuce plants subjected to varying irrigation regimes under controlled conditions. Four water availability treatments were imposed − Full Control (FC), Stress Control (SC), Mild Stress (SM), and Severe Stress (SS) − varying in timing, frequency, and intensity of irrigation protocols. Using dense optical flow on time-lapse RGB images, we extracted MK features that link leaf age to motion dynamics. These high-dimensional temporal features were compressed into descriptive and trend-based characteristics for classification. Multi-classification problem was divided into nine sub-tasks, for which feature selection and multiple machine-learning models were tested applying Leave-One-Sample-Out cross-validation. The best models were organised into four explainable hierarchical cascades. The presented system captures enough information to successfully distinguish among subtle differences in plants’ response to water availability dynamics (best architecture cascade obtained 0.93 out of fold balanced accuracy). The framework associating leaf age with MK features along with feature engineering allowed explainability – e.g., central rosette’s features were selected almost twice the expected frequency (19 out of 58) in tasks involving the stress-adapted control (SC), while features capturing linear trends in motion were generally selected over twice as often as simple descriptive statistics (44 vs. 19), proving essential for distinguishing most stress conditions. The MK approach proved effective for differentiating water stress levels, positioning it as a powerful tool for digital phenotyping and a solid foundation for developing advanced temporal-aware models.
Why it matches plant phenotyping methods画像時系列とdense optical flowから植物のモルフォ・キネマティック形質を抽出し、水分状態を分類する手法の開発・検証が研究の中心であるため。
abstractwe present a structured image-based approach to define and test morpho-kinematic (MK) traits from lettuce plants
CottonAerial / UAVLiDAR / point cloudMultispectral / hyperspectralThermalLeafPhysiological trait estimation2D/3D reconstructionImage / point-cloud registrationWater status / transpiration
Accurate estimation of crop water status is essential for monitoring plant senescence and enabling intelligent agricultural management. This study proposes a pixel-aligned co-registration and DSM-grid fusion framework that integrates high-resolution point clouds, multispectral (MS) images, and thermal imagery acquired by Unmanned Aerial Vehicles (UAVs) to enable three-dimensional prediction and visualization of cotton canopy leaf water content (LWC) and equivalent water thickness (LEWT). To address the low spatial resolution of thermal imagery, a downsampling–upsampling simulation framework was developed to evaluate interpolation errors. This framework quantitatively compares three common interpolation methods—nearest neighbor, bilinear, and bicubic interpolation—using RMSE and PSNR metrics. Results show that bicubic interpolation performs best in preserving spatial details and minimizing errors, and is therefore adopted in the subsequent image fusion process. A 3D grid was constructed based on the digital surface model (DSM), enabling grid-cell (pixel-aligned) spectral and thermal features to be mapped onto point-cloud units. Vegetation and thermal indices extracted from the mapped features were used as input variables. Combined with recursive feature elimination (RFE) and random forest (RF) models, the prediction of LEWT and LWC achieved R² values of 0.792 and 0.752, and rRMSE values of 13.84% and 9.68%, respectively. These results significantly outperformed those of partial least squares regression (PLSR), support vector machine (SVM), and extreme learning machine (ELM) models. By integrating the predicted results with the point cloud data, a 3D representation of canopy water parameters was constructed, revealing a typical top-down gradient of water loss. The experiment also revealed that nitrogen treatment significantly influenced the vertical distribution of water content. High-nitrogen application delayed water loss in the middle and lower canopy layers, highlighting the coupled regulation between nitrogen and water. Parameter comparisons showed that LEWT exhibited higher sensitivity than LWC across both temporal and spatial scales, making it a more robust indicator for canopy water monitoring. Additionally, point clouds generated from Cross-circling oblique (CCO) photogrammetry outperformed UAV LiDAR systems in terms of point density, structural completeness, and image fusion potential. In summary, this study validated the feasibility and effectiveness of integrating point cloud, MS, and thermal imagery via the proposed pixel-aligned co-registration and DSM-grid fusion framework for 3D crop water monitoring. The proposed method provides a reliable technical foundation for drought detection, irrigation management, and yield prediction in precision agriculture.
Why it matches plant phenotyping methodsUAVマルチスペクトル・熱画像・点群を統合し、綿花キャノピーの水分形質を3D推定・可視化する手法の開発と検証が中心である。
abstractThis study proposes a pixel-aligned co-registration and DSM-grid fusion framework that integrates high-resolution point clouds, multispectral (MS) images, and thermal imagery acquired by Unmanned Aerial Vehicles (UAVs) to enable three-dimensional prediction and visualization of cotton canopy leaf water content (LWC) and equivalent water thickness (LEWT).
Plant disease adversely impacts food production and quality. Alongside detecting the disease, estimating its severity is important in managing the disease. Artificial intelligence deep learning-based techniques for plant disease detection are emerging. Unlike most of these techniques, which focus on disease recognition, this study addresses various plant disease-related tasks, including annotation, severity classification, lesion detection, and leaf segmentation. We propose a novel approach that learns the disease symptoms, which are then used to segment disease lesions for severity estimation. To demonstrate the work, a dataset of barley images was used. We captured the images of barley plants inoculated with diseases on test-bed paddocks at various growth stages. The dataset was automatically annotated at a pixel level using a trained vision transformer to obtain the ground truth labels. The annotated dataset was applied to train salient object detection (SOD) methods. Two top-performing lightweight SOD models were used to segment the disease lesion areas. To evaluate the performance of the SODs, we have tested them on our dataset and several other datasets, including the Coffee dataset, which has expert pixel-level labels that were unseen during the training step. Several morphological and spectral disease symptoms, including those akin to the widely used ABCD rule for human skin-cancer detection, i.e., asymmetry (A), border irregularity (B), colour variance (C), and diameter (D), are learned. To the best of our knowledge, this is the first study to incorporate these ABCD features in plant disease detection. We further extract visual and texture features using the grey level co-occurrence matrix (GLCM) and fuse them with the ABCD features. For the coffee dataset, our method achieved 82+% detection accuracy on the severity classification task. The results demonstrate the performance of the proposed method in detecting plant diseases and estimating their severity.
Why it matches plant phenotyping methods植物病害の病変を画像から抽出し、病害重症度という植物状態を推定する手法の開発・評価が中心であるため。
titleAutomatic pixel-level annotation for plant disease severity estimation
High-throughput field phenotyping bridges genotype, environment, and phenotypic performance. Conventional plot-level approaches relying on manual surveys are labor-intensive, and error-prone and fail to capture variability among individual plants, limiting seed cotton yield estimation and genotype screening under natural conditions. To address these limitations, a complex framework was developed, integrating single-plant instance segmentation, multi-trait inversion, plot-level stability characterization, and yield estimation. The enhanced vision-model framework, TopoRefineSAM, combines YOLOv12 detection with SAM2 segmentation and incorporates adaptive enhancement and topological refinement modules, enabling efficient, robust, and cost-effective single-plant identification under weak annotation. Based on this segmentation, multi-source UAV imagery (RGB, multispectral, thermal infrared, and DSM) was used to build ensemble learning models for inversion of physiological and biomass traits. A Stability Index Group (SIG) translates inter-plant variability into plot-level stability features, improving interpretability and consistency in yield estimation and cultivar screening. Results demonstrated that TopoRefineSAM achieved high segmentation accuracy for single-plant extraction under complex field conditions. In multi-trait inversion, Gradient Boosting Decision Trees (GBDT) achieved the highest performance. Our results demonstrated strong consistency between multimodal features and measured traits. In yield estimation, incorporating the SIG substantially improved predictive performance across growth stages. In cultivar screening, the method achieved high agreement with field measurements, showing robust identification of top-performing cultivars. Collectively, the findings establish a scalable, cost-effective, and high-accuracy framework for field-based phenotypic analysis and yield estimation, providing both methodological innovations and practical support for precision breeding and large-scale crop improvement.
Why it matches plant phenotyping methods単一植物のセグメンテーション、マルチモーダル画像による形質推定、安定性指標、収量推定を統合した圃場フェノタイピング手法の開発が中心である。
abstractTo address these limitations, a complex framework was developed, integrating single-plant instance segmentation, multi-trait inversion, plot-level stability characterization, and yield estimation.
AppleField / plotLiDAR / point cloudRGB-D / ToFFruitObject detection2D/3D reconstructionSegmentation
One of the key challenges in orchard robots is accurately localizing occluded fruits in complex environments, especially when the fruit targets are split into multiple isolated regions within images. Traditional single-task network models exhibit limited capability in discerning fragmented targets that belong to the same fruit but are segmented into multiple spatially isolated regions within images. In addition, fruit localization largely relies on high-cost sensors or additional 3-D localization algorithms. To address this issue, we propose a fruit detection and centroid localization method based on a Multi-Task Wavelet-Enhanced YOLO (MT-WavYOLO) to enhance the success rate of robotic operations on occluded fruit targets. Initially, a lightweight semantic segmentation branch was integrated into the YOLOv8 backbone network to precisely segment exposed fruits, while retaining the original object detection branch to fully identify occluded fruits. To address the diminished sensitivity of conventional models to geometric profiles of heavily occluded fruits, a novel feature fusion module, C2f_WTConv, was designed by incorporating wavelet transform convolution, leveraging the multi-frequency robustness of wavelet representations to enhance the model’s feature extraction capabilities under complex orchard occlusions. Subsequently, a 3D frustum-based point cloud processing method was proposed, combining the detection results from MT-WavYOLO with the semantic segmentation masks to accurately localize occluded fruits. MT-WavYOLO demonstrated a 2%, 1.5%, and 2.2% improvement in Precision, Recall, and mAP50, respectively, on our custom-built dataset compared to the latest YOLOv10s model. Semantic segmentation performance, measured by Intersection over Union (IoU) and Accuracy, was improved by 5.2% and 3.8%, respectively, over the state-of-the-art Deeplabv3+ network. Compared to the adapted multi-task network YOLOP, MT-WavYOLO achieved a 3.4% increase in mAP50 and a 2.7% improvement in IoU. In addition, MT-WavYOLO has a compact footprint of 10.2 M parameters and achieves approximately 27 FPS in real-time inference, thereby meeting the requirements of robotic harvesting operations. The proposed localization method was evaluated through 600 fruit localization tests using six different RGB-D cameras in an orchard environment. The average experimental results demonstrated that the centroid localization and radius estimation errors were reduced by 42.5%, 73.7%, 16.17%, and 11.25%, respectively, compared to traditional 3D bounding box methods and our previous approaches. These results indicate that the MT-WavYOLO combined with the frustum-based method significantly enhances the accuracy of apple localization under complex orchard conditions using consumer-grade sensors, providing a strong practical foundation for non-destructive robotic harvesting.
Why it matches plant phenotyping methods果実の検出・3D重心定位という植物器官の形態的状態を、画像分割・深層学習・点群処理で推定する手法を開発し、データセットおよび複数カメラで性能評価しているため、方法が中心的である。
abstractwe propose a fruit detection and centroid localization method based on a Multi-Task Wavelet-Enhanced YOLO (MT-WavYOLO)
Aiming at the hysteresis problem of traditional contact-type mechanical throughput detection method during combine harvester operation, this study proposes a multimodal data-driven online throughput prediction method. By building a multimodal sensor system that integrates vehicle-mounted cameras, GPS, grain moisture content, and feeding auger power sensors, a throughput prediction framework based on wheat ear biomass characteristics was established: Firstly, the WEC-MVFF wheat ear online counting density map estimation model is designed, and MobileViT is used to build a feature extraction backbone network. The multi-scale fusion module and the centralized conversion module are combined to realize the collaborative extraction of shallow texture features and deep semantic features of dense wheat ears in the field. Secondly, divide the image into regions of interest and establish a throughput prediction model based on multimodal information of wheat ear number (image) − moisture content (sensor) − travel speed. At the same time, a detection model based on feeding auger power is constructed as a comparison benchmark. Field tests show that the WEC-MVFF model maintains an average counting accuracy of more than 90 % under different wheat ear density, travel speed and light intensity conditions. The model’s online counting advantage is verified through ablation study and comparative tests with other counting models. The throughput prediction method achieved an MAE of 0.70 kg/s and 0.77 kg/s in test area 1 and 2 respectively, the first-order difference fluctuation was stable in the range of ±0.50 kg/s, and the prediction frame rate of 10-13fps met the real-time requirements. Compared with the single-modal image prediction method, the accuracy of the multimodal method considering moisture content was improved by 0.80 kg/s and 0.75 kg/s in the two test areas, respectively. Compared with traditional mechanical quantity detection methods, it has higher accuracy and stability while achieving early prediction, providing reliable feedforward information support for the intelligent control of harvesters.
Why it matches plant phenotyping methods小麦穂の画像計数とマルチモーダルセンサを用いて、穂密度・バイオマス特性および収量流量を推定する手法を開発・検証しており、植物形質の取得・推定が研究の中心である。
Chlorophyll content is an important indicator of rice growth status. Accurately estimating the nutritional status of rice canopies using hyperspectral data and inversion models is of great significance for precision farming. Ground-based spectrometers can obtain precise spectral data, but they are limited by spatial resolution and cannot be used for large-scale observations. Unmanned aerial vehicle (UAV) imaging spectrometers can be used to observe large areas of farmland, but due to sensor limitations, the data accuracy is poor, which in turn leads to poor inversion accuracy. This study optimizes UAV hyperspectral data based on ground-based spectrometer data. A wavelength random combination traversal algorithm is used to select wavelengths. Inversion models for rice chlorophyll content are constructed using ELM, CPO-ELM, and FLA-ELM. The results indicated that the optimized hyperspectral reflectance was highly consistent with ASD reflectance. The root mean square error of reflectance (RMSEReflectance) across all bands decreased from 0.108 to 0.012 (88.89 % reduction), and the average RMSEReflectance across all samples decreased by 88.64 %. Compared to the original UAV data, the optimized data achieved the best inversion performance with the FLA-ELM model: the coefficient of determination (R²) of the test set increased from 0.608 to 0.755 (24.2 % relative improvement), and the RMSE of chlorophyll content (RMSEChl) decreased from 6.518 μg/cm2 to 5.371 μg/cm2. These results validate the effectiveness of the proposed spectral optimization scheme. In conclusion, modeling based on optimized UAV hyperspectral imagery improves the accuracy of rice chlorophyll content inversion, providing a novel approach for rice nutritional monitoring.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像からイネのクロロフィル含量を推定する手法を最適化し、精度を定量的に検証しており、植物フェノタイピング手法が中心である。
abstractThis study optimizes UAV hyperspectral data based on ground-based spectrometer data.
Powdery mildew is a major disease affecting rubber tree yield. Rapid and accurate monitoring of this disease is crucial for plantation management. Previous studies have focused on spectral and spatial data for monitoring powdery mildew but have not adequately addressed the underlying physiological and biochemical alterations induced by the disease. Therefore, this study proposes a method combining spatial-spectral features and plant traits to monitor rubber tree powdery mildew. Unmanned Aerial Vehicles (UAVs) equipped with the ULTRIS X20P hyperspectral sensor (350–1000 nm) were used to capture hyperspectral imagery in two rubber plantations. Plant traits (PTs), including chlorophyll (Cab), carotenoids (Car), anthocyanins (Anth), leaf water content (Cw), and dry matter content (Cm), were inverted from UAV hyperspectral imagery using a radiative transfer model. Meanwhile, spectral and spatial information in the imagery were analyzed to extract vegetation indices (VIs), texture features (TFs), and color features (CFs) sensitive to the disease. Machine learning algorithms, including Partial Least Squares Regression (PLSR), Least Absolute Shrinkage and Selection Operator (LASSO), and Random Forest Regression (RF), were subsequently employed to create disease monitoring models based on these features. The results show that models based on VIs and TFs effectively monitor powdery mildew, with the inclusion of PTs significantly improving model performance. Models that integrate multiple features outperform those that depend on single features, especially the monitoring model integrating VIs, TFs, and PTs using the PLSR algorithm, which achieved an R² of 0.794 and an RMSE of 7.991. This study highlights the novelty of integrating spatial-spectral features with plant traits for monitoring rubber tree powdery mildew, offering a reference for precise disease monitoring through the use of UAV hyperspectral imagery.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像から植物形質を推定し、空間・スペクトル特徴と統合して病害症状を監視する手法が研究の中心であるため。
abstractTherefore, this study proposes a method combining spatial-spectral features and plant traits to monitor rubber tree powdery mildew.
Accurate and efficient estimations of leaf area index (LAI) are crucial for crop management, including intelligent crop breeding, nutrient management, and yield prediction. Unmanned aerial vehicle (UAV)-based multispectral sensors with machine learning models provide high-precision solutions for LAI estimation but are hindered by the challenge of acquiring adequate ground truth data. Transfer learning, one type of deep learning framework, offers a solution by leveraging prior knowledge learned through models that have been pre-trained, thus ensuring robust performance despite limited data availability. This study evaluates the efficacy of the fine-tuning PROSAIL-Informed Deep Neural Network model (PROSAIL-DNN) for estimating LAI of different oat varieties and growth periods using UAV multispectral images. We compared this model with several widely used machine learning algorithms, including partial least squares (PLS), least absolute shrinkage and selection operator (Lasso), support vector regression (SVR), and extreme gradient boosting (XGBoost)), and a DNN, using training datasets consisting of 70 %, 60 %, and 50 % of the field data to assess the impact of data volume on model performance. Our results demonstrated that the PROSAIL-DNN model outperformed other algorithms across different growth periods and all growth periods. Specifically, with only 50 % of the training data used, the average R² of the PROSAIL-DNN model was 5.44 %, 28.76 %, 12.98 %, 12.72 %, and 22.90 % higher than those of DNN, Lasso, PLS, SVR, and XGBoost, respectively. The PROSAIL-DNN model also demonstrated higher accuracy in monitoring LAI during different growth periods, especially at early jointing period (P1) (R² = 0.838, RMSE = 0.376, and RPD = 2.483) and at post-heading period (P3) (R² = 0.881, RMSE = 0.298, and RPD = 2.896). Our findings underscore the potential of combining PROSAIL with deep transfer learning to accurately and robustly estimate oat LAI at various growth periods using UAV multispectral images with limited field data. This approach provides a solid foundation for applying transfer learning in crop monitoring and can be adapted for other crop variables in future studies.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像とPROSAIL-DNNを用いて、オート麦の葉面積指数（LAI）を推定する画像解析・機械学習手法が研究の中心であり、複数モデルとの技術比較とデータ量による性能評価も実施している。
abstractThis study evaluates the efficacy of the fine-tuning PROSAIL-Informed Deep Neural Network model (PROSAIL-DNN) for estimating LAI of different oat varieties and growth periods using UAV multispectral images.
Canopy nitrogen density (CND) is a critical indicator of plant growth, with applications in nutrient diagnosis, disease monitoring, and carbon cycling. However, optical remote sensing of nitrogen is constrained by VI saturation, particularly in dense canopies. Here, we propose a novel strategy to mitigate saturation by resolving two key issues: (1) the sensing depth of canopy spectra and (2) the quantification of vertical nitrogen heterogeneity. Saturation characteristics of VIs were first analyzed using inflection and saturation points. We found that CND at the inflection point enhanced the linear correlation with canopy spectra. The 3rd–7th leaf layers contributed most to canopy reflectance, accounting for 67.24%–72.15% of canopy spectra and 62.39% of total CND. Beyond the 7th–8th leaf layers, saturation became prominent. To capture vertical heterogeneity, we employed a bell-shaped model, with the coefficient ω linking the inflection point CND to canopy CND across structural variations. Integrating inflection points with vertical heterogeneity characteristics improved the robustness of VI–CND relationships, reducing RMSE by 10.8%–33.9%. This approach offers an intuitive framework to mitigate VI saturation, enabling more accurate CND estimation under diverse field conditions.
Why it matches plant phenotyping methods光学リモートセンシングによる作物キャノピー窒素密度の推定法を開発し、飽和効果と垂直的不均一性を補正して精度を検証しているため、植物形質取得が研究の中心である。
abstractHere, we propose a novel strategy to mitigate saturation by resolving two key issues: (1) the sensing depth of canopy spectra and (2) the quantification of vertical nitrogen heterogeneity.
Flow disturbances induced by mechanical weeding significantly affect the mechanical behavior of rice seedlings, posing a critical challenge to the mechanization of paddy field operations. To elucidate the mechanisms by which weeding blades affect the mechanical properties of rice seedlings, we developed a coupled CFD model integrating air–water–mud three-phase flow with the flexible structure of rice seedlings. The model accounts for both hydrodynamic forces and the biomechanical properties of seedlings, enabling accurate prediction of deflection, displacement, and stress responses under blade-induced disturbance. Validation experiments, including seedling deflection under steady flow and field measurements of flow fields around operating blades, demonstrated that the model’s predictions deviate from measured data by less than 10 %, confirming its accuracy and robustness. The results indicate that, during operation, maximum seedling stress reached 8.37 MPa, and maximum intra-row displacement was 53.34 mm (≈ 20 % of seedling height) at 10 days after transplanting, decreasing by 29.9 % by 30 days as stiffness increased. Stress concentration occurred primarily at the seedling base and near the water–air interface, indicating critical regions for structural failure. These findings provide new mechanistic insight into the coupled dynamics of seedlings, fluid, and weeding blades, establishing a quantitative foundation for optimizing blade spacing, rotational speed, and working depth to minimize seedling damage during mechanical weeding.
Why it matches plant phenotyping methodsイネ苗の変位・たわみ・応力という植物状態を推定する連成CFDモデルを開発し、実測値で検証しており、植物表現型の取得・推定手法が研究の中心である。
abstractwe developed a coupled CFD model integrating air–water–mud three-phase flow with the flexible structure of rice seedlings
Improving nitrogen use efficiency (NUE) in commercial maize production remains a persistent challenge. A major barrier is the lack of simple, remote sensing–based decision-support frameworks that enable broad adoption of in-season, site-specific nitrogen (N) management. This study developed and evaluated a practical framework that contextualizes the Holland–Schepers sensor algorithm using PlanetScope (PS) satellite imagery to guide multiple in-season, variable-rate fertigation delivered through a flow-proportional injection system integrated with a center pivot system equipped with variable-rate irrigation. Field implementation was carried out during the 2023 and 2024 seasons across four N rates (0-N, Low-N, Fertigation, Full-N) and three irrigation treatments: full (BMP), deficit (50 %BMP), and rainfed. Normalized Difference Red Edge (NDRE)-derived sufficiency index (SI) values informed the amount, timing, and spatial distribution of N applications. In 2024, PS-guided fertigation achieved yields statistically comparable to Full-N while reducing total N input by 23 %. Significant improvements in NUE were observed, with Fertigation outperforming Full-N by 12 % in agronomic efficiency (AE) and 26 % in partial factor productivity of N (PFPN). Satellite-derived SI values were strongly correlated with UAV benchmarks from the MicaSense Altum and RedEdge-3 sensors (ρ = 0.82–0.95). However, PS consistently overestimated NDRE relative to UAV data, particularly under N-deficient conditions, underscoring the need for local calibration and bias correction. To improve diagnostic specificity, a biologically informed, rule-based stress-classification framework was developed to differentiate nitrogen stress from water stress using NDRE and soil water depletion (SWD) as diagnostic variables. Retrospective yield and management data were used to establish physiologically meaningful NDRE–SWD thresholds for stress diagnosis during the critical in-season fertigation window (V10–R2). Full-Yield plots achieved approximately 12,000 kg/ha at NDRE = 0.78 and SWD = 64 mm. The resulting NDRE–SWD–yield patterns highlight the feasibility of disentangling stress types under commercial field conditions. However, further validation across seasons and environments, along with integration of canopy water or temperature indices, is needed to improve water-stress detection and enable real-time decision-making. Overall, these results offer actionable guidance for implementing satellite-guided fertigation at commercial scale. The developed framework delivers scalable, data-driven N recommendations that enhance profitability and support environmentally responsible maize production. It also provides a reference for future research and extension programs aiming to turn satellite remote sensing into practical tools for site-specific N management.
Why it matches plant phenotyping methods衛星・UAVリモートセンシングによるNDRE指標を用いて植物の窒素・水ストレスを診断し、センサー間の検証とルールベース分類法の開発を行っているため、植物状態の取得・推定手法が中心的である。
abstractThis study developed and evaluated a practical framework that contextualizes the Holland–Schepers sensor algorithm using PlanetScope (PS) satellite imagery to guide multiple in-season, variable-rate fertigation
Maize ear traits are critical indicators for elucidating yield formation mechanisms and are widely used in genetic studies. Traditional two-dimensional (2D) phenotyping suffers from planar analysis constraints, occlusions in single-view imaging, and limited robustness to mixed textures or curved ears. To address these issues and support germplasm archiving and breeding research, we developed MaizeEar3DPheno (MEP3D), a 3D point cloud-based method for quantifying maize ear phenotypic traits. A structured-light 3D scanning system equipped with a motorized rotary platform was designed to acquire point clouds of 30 maize ears from three different varieties. Preprocessing involved axis alignment via PCA, uniform downsampling, and removal of non-kernel regions. Following preprocessing, key phenotypic traits, including ear length, diameter, and barren tip length, were calculated from the processed point clouds. MEP3D integrated directional erosion with density-based clustering to achieve robust kernel segmentation and counting. A spatial analysis algorithm was further developed to locate kernel row arrangements from geometric features. The results demonstrated that the proposed method achieved high-precision cross-variety kernel counting, with a mean absolute percentage error (MAPE) of 0.91%, and a coefficient of determination (R²) of 0.9917 across all maize ears. Kernel row quantification was fully consistent with manual measurements, allowing extraction of row inclination and average kernel number per row. Ear length, diameter, and barren tip length estimation achieved R² values of 0.9864, 0.9871, and 0.9670, respectively, demonstrating robustness. The generated high-fidelity 3D phenotypic data supports automated evaluation of ear and kernel traits and facilitates in-depth analysis of spatial morphological characteristics.
Why it matches plant phenotyping methodsトウモロコシ雌穂の3D点群取得・処理・形質抽出法を開発し、カーネル数や穂長などを手動測定と比較検証しており、フェノタイピング手法が研究の中心である。
abstractwe developed MaizeEar3DPheno (MEP3D), a 3D point cloud-based method for quantifying maize ear phenotypic traits.
Accurate monitoring of crop phenology, biophysical attributes and agroclimatic variability is essential for optimizing agricultural practices, particularly in smallholder farming systems. In this study, we evaluated how field camera-derived Green Chromatic Coordinate (GCC) reflected variations in agroclimatic factors (rainfall and soil moisture) and biophysical attributes (leaf area index, crop height, chlorophyll content, and stomatal conductance) across agroecological zones (AEZs) in Kenya. Next, we utilized GCC time series to detect six key phenological stages of maize (Zea mays L.) - emergence, stem elongation, tasseling, kernel development, ripening, and senescence - using an amplitude-based relative threshold method. This approach was cross-validated against field observed phenology. Our analysis revealed positive correlations between GCC and plant height, chlorophyll content, and leaf area index (LAI). Daily-scale Pearson lag correlation between GCC and agroclimatic factors revealed that crops in drier ecosystems exhibited shorter response times to agroclimatic fluctuations (32 days to rainfall and 14 days to soil moisture), highlighting site-specific differences in vegetation dynamics captured by field cameras. Furthermore, results indicate that GCC effectively captured phenological stages with high accuracy (R² = 0.9, RMSE = 7.1–7.7 days), though variability was observed across sites and growth stages. Comparisons between within-site and inter-site validation suggest that localized calibration can improve accuracy. Nevertheless, the method remains robust across varying conditions, which is supported by comparison against established curve-fitting methods. Our findings highlight the potential of field cameras as a cost-effective tool for crop monitoring at a high spatial and temporal scale, with applications in crop phenology detection, biophysical monitoring, and validation of remote sensing products. Integrating this approach into regenerative agriculture frameworks could enhance decision-making and management interventions in smallholder farms.
Why it matches plant phenotyping methods圃場カメラ画像からGCCを抽出し、トウモロコシの生育ステージと生物物理形質を推定する手法を開発・検証しており、フェノタイピング手法が研究の中心である。
abstractwe utilized GCC time series to detect six key phenological stages of maize
Efficient and accurate extraction of plant height (PH) plays an important role in analyzing its deeper phenotypic traits and improving breeding efficiency. Traditional methods make it difficult to measure PH at the individual plant level in the plot-level on a large scale, with high accuracy and low delay. To address this issue, we adopted a truss-type phenotyping platform to acquire LiDAR and RGB canopy data from 120 rapeseed genotypes from the two-leaf stage to the flowering stage, covering a total of 6 growth stages. (1) Object detection was performed to identify per plant of rapeseed images by the K-Means improved Faster R-CNN algorithm. (2) The image data before and after object detection and the point cloud data were fused to recognize per plant on the point cloud. Besides, the rapeseed plant point cloud and the ground point cloud were distinguished by color. (3) The Cloth Simulation Filter (CSF) algorithm is used to fit the ground points obscured by the canopy, which contributes to accurately extracting the PH of individual rapeseed. The field tests indicated that the mAP (IoU = 0.5) of the improved object detection method was 0.902, which achieved a high detection accuracy for rapeseed with different sizes in all periods. Specifically, in the PH accuracy verification of the No. 077 cultivar, R² was 0.997, RMSE was 1.156 cm, rRMSE was 5.48 %, and the maximum difference between automatic recognition and manual measurement in the late stage of growth was less than 6 cm. Especially, the proposed method can extract of the height of individual plant with the values of R² was 0.980, RMSE was 0.651 cm, rRMSE was 6.543 % in the seedling stage and the PH was lower than 20 cm. The differences of pH of 40 rapeseed genotypesunder cold stress were compared, which provide a reference for high-throughput PH extraction and exploring genotypic differences in plant breeding.
Why it matches plant phenotyping methodsLiDAR・RGBデータ融合と画像／点群処理により、個体ごとの草丈を高精度・ハイスループットに抽出する手法の開発と検証が中心であるため。
titleHigh-throughput extraction of individual plant height in rapeseed based on LiDAR-Camera data fusion
Agricultural productivity is the major source of the economy of the agrarian nation and is also considered the major source of human activities. It is vital to detect and manage plant diseases to enhance the overall growth and quality of agricultural produce. Modern agriculture increasingly uses IoT and automation tools to monitor and collect valuable information on plant health and disease occurrence. Here, the plant disease is detected using Deep Recurrent Visual Geometry Group-16 (DRVGG-16) in the IoT platform with routing based on the Adam Dung Beetle Optimization (ADBO) algorithm. The DRVGG-16 is the combination of Deep Recurrent Neural Network (DRNN) with Visual Geometry Group-16 (VGG-16) and ADBO is the fusion of Adam optimizer with Dung Beetle Optimization (DBO). Initially, the captured agricultural images are transmitted to the base station via ADBO-optimized routing for plant disease detection. The base station initiates disease detection by applying a bilateral filter to preprocess the input plant leaf images. The preprocessed images are segmented using the DeepLabV3+ model, and features are subsequently derived from the segmented leaf areas. Finally, plant disease detection is performed using the proposed DRVGG-16 model. Furthermore, the performances of the proposed schemes are validated, where ADBO attained distance, delay and energy of 0.343 m, 0.305 ms and 0.375 J. The DRVGG-16 obtained superior results of 91.43 % accuracy, 90.63 % True Positive Rate (TPR), 90.73 % precision, 92.34 % True Negative Rate (TNR) and 90.68 % F1-score. Therefore, the proposed IoT-based framework combining DRVGG-16 and ADBO routing effectively detects plant diseases with high accuracy while optimizing energy and transmission efficiency.
Why it matches plant phenotyping methods植物葉画像から病害状態を推定する画像解析手法を提案し、セグメンテーションと分類モデルの性能を検証しており、病害フェノタイピング手法が研究の中心である。
abstractHere, the plant disease is detected using Deep Recurrent Visual Geometry Group-16 (DRVGG-16) in the IoT platform with routing based on the Adam Dung Beetle Optimization (ADBO) algorithm.
Accurate segmentation and analysis of root images from soil-grown plants are critical for advancing our understanding of root growth and plasticity under varying environmental conditions. Most approaches typically rely on binary segmentation of the entire root system architecture (RSA), which limits their ability to capture the hierarchical complexity of root structures, including axial and lateral roots. To address this, our study evaluated five convolutional neural network (CNN) architectures for multi-class (i.e. axial/lateral) segmentation of 2D root images from barley plants grown in rhizoboxes: (i) U-Net, (ii) U-Net with Atrous Spatial Pyramid Pooling (UnetASPP), (iii) U-Net with Attention Block (UnetAtt), (iv) DeepLabV3+ with MobileNetV2 (DLMB), and (v) DeepLabV3+ with ResNet-50 (DLR50). Among these, the DLR50 model achieved the highest segmentation accuracy, particularly for distinguishing lateral roots within complex RSA structures. Furthermore, analysis of root traits derived from the segmented images confirmed that DLR50 produced the most reliable estimations of phenotypic traits compared to ground truth measurements. These findings highlight the strong potential of advanced multi-class CNN models—especially DLR50—for detailed and quantitative analysis of soil-root systems, providing new insights into root responses to environmental conditions.
Why it matches plant phenotyping methods根画像から軸根・側根を分割し、分割画像に基づく表現型形質推定のCNN手法を開発・比較検証しており、植物フェノタイピング手法が研究の中心です。
abstractour study evaluated five convolutional neural network (CNN) architectures for multi-class (i.e. axial/lateral) segmentation of 2D root images from barley plants grown in rhizoboxes
Accurately estimating the number of cotton bolls is vital for plant phenotyping, offering essential insights for both breeders and growers. This trait offers valuable phenotypic information on plant productivity and supports crop management decisions to optimize yield and profitability for growers. Manual counting of bolls in the field, however, is impractical because it is labor-intensive and time-consuming. This study presented a video-based cotton boll counting approach that integrated a transformer-based detector (RT-DETR) with multi-object tracking techniques. To prevent double-counting bolls across frames, two motion estimation methods, FlowFormer and TAPIR were explored to predict the movement of bolls between adjacent frames and a two-stage association process combining Intersection over Union (IoU) and Euclidean distances was developed to track bolls across time. To further enhance counting accuracy, a virtual counting line was introduced to reduce ID switch errors. Experimental results demonstrated the effectiveness of the RT-DETR model, achieving an mAP0.5 exceeding 0.93 for dense boll detection. Furthermore, both FlowFormer and TAPIR can be used for tracking cotton bolls in the videos while the tracking performance of the FlowFormer-based method was slightly higher than that of the TAPIR-based method with an MOTA of 73.36 % and an IDF1 of 79.89 %. The tracking approach integrating RT-DETR and FlowFormer exhibited a relatively strong correlation between the predicted and the ground-truth boll number with an R² of 0.60 and an MAPE of 14.34 % on multi-plant plots. In single-plant plots, the approach achieved a high correlation with an R² of 0.97 and a MAPE of 10.33%. These findings indicated the potential of the proposed approach as an effective, automated tool to support breeding programs and yield assessments in cotton production. Both the code and dataset can be accessed at: https://github.com/UGA-BSAIL/Dense_cotton_boll_counting.
Why it matches plant phenotyping methods綿花のボール数という植物生産形質を、動画検出・追跡とロボット収集で自動推定する手法の開発・評価が研究の中心であり、mAP、MOTA、IDF1、R²、MAPEによる技術検証も行っている。
abstractThis study presented a video-based cotton boll counting approach that integrated a transformer-based detector (RT-DETR) with multi-object tracking techniques.
Three-dimensional (3D) reconstructions of orchards offer richer data for digital phenotyping and underpin smart-agriculture applications. However, achieving high-level reconstruction quality and robustness is challenging due to the complex structure of the orchard. This study presents a novel framework that provides centimeter-level 3D realtime reconstructions and phenotyping for apple orchards. A multi-RGBD camera array was adopted, creating a wide overlapping view and robust features. The hybrid odometry front-end and the dual loop-closure strategy ensured low-drift pose estimation. The generated pointcloud was input into the ellipsoid-fitting routine to extract fruit diameter and volume. We validated this framework through reconstruction and phenotypic errors in four rows of an apple orchard with different tree spacings. The root mean square error of the absolute trajectory error in global reconstruction was less than 16 mm. The mean absolute percentage error (MAPE) of the local fiducial distance of approximately 5 m was less than 0.12%. The system was implemented at higher than 12.5 frames per second in an embedded system. The MAPEs of the fruit’s diameter were 2–2.17%, and those of its volume were 5.3–5.6%. Additionally, ablation experiments were carried out on multi-camera and loop-closed elements, and comparisons were made with existing methods to further demonstrate their effectiveness. In conclusion, this research provides an efficient and stable deployable solution for 3D reconstruction of orchards, which is conducive to the development of more advanced and multilayer modern orchard models and promotes the practice of smart agriculture.
Why it matches plant phenotyping methodsリンゴ園向けのマルチRGB-Dによる3D再構成と、点群から果実径・体積を抽出するフェノタイピング手法を開発・検証しており、方法が研究の中心である。
abstractThis study presents a novel framework that provides centimeter-level 3D realtime reconstructions and phenotyping for apple orchards.
Monitoring plant weight during its development cycle is crucial for effective growth monitoring; it provides valuable information on plant’s health and development. Weight data are essential to determine the optimal harvest time and ensure that plants are harvested when they are at their best. This work aimed to design and implement a workflow that allows the study of growth and development variables in a spinach crop cycle using high spatial resolution multispectral images acquired with an Unmanned Aerial Vehicle (UAV). We based this workflow on applying a multitask attention U-Net model for plant segmentation and advanced statistical analysis, including regression methods and hierarchical analysis, to build and evaluate a Random Forest (RF) model and a Generalized Linear Model (GLM) for spinach fresh weight estimation. The segmentation model achieved a mean Intersection over Union (mIoU) of 0.90 and an F-score of 0.93 against manually drawn labels. Experimental validation of the estimation of the fresh weight of spinach plants from geometric and spectral characteristics with an R2 of 0.90 and RMSE = 23.48 g for the RF model constructed from explanatory variables found with hierarchical analysis. Results demonstrate the utility of a novel hybrid approach for analysis of multispectral imagery from UAVs in crop monitoring.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像から植物セグメンテーションと幾何・スペクトル特徴を抽出し、ホウレンソウの生体重を推定するワークフローを開発・検証しており、表現型取得・推定手法が中心である。
abstractThis work aimed to design and implement a workflow that allows the study of growth and development variables in a spinach crop cycle using high spatial resolution multispectral images acquired with an Unmanned Aerial Vehicle (UAV).
Addressing the challenges of variable target morphology, small critical regions, and complex background interference in eggplant picking point detection within complex agricultural scenarios, this study proposes MDAD-YOLO (Multi-dimensional Attention and DySample YOLO), a detection model improved based on the YOLOv10n-pose framework. First, the model’s cross-dimensional perception ability for fruits and picking points is enhanced by integrating the collaborative mechanism of regional receptive field attention with channel-space joint attention. Next, within the Neck structure, coordinate attention is incorporated to optimize the spatial localization accuracy of fine-grained features, enhancing sensitivity to minute regions such as the fruit stem apex. Additionally, dynamic pixel reorganization is applied to enhance feature map reconstruction details, addressing the detail loss caused by traditional interpolation methods. Finally, cascading adaptive fine-grained channel attention with position-sensitive attention enables multi-level modeling of channel dependencies and collaborative spatial context enhancement. Through a seven-tier validation framework, the model’s effectiveness, robustness, and generalizability have been comprehensively demonstrated. Experimental results show that the model achieves 93.6% mAP@50 for object detection, 94.7% mAP@50 and 92.1% mAP for keypoints detection, and an average pixel Euclidean distance error of 19.41 on the self-built eggplant dataset, outperforming YOLOv12 and other high-performance models. Additionally, cross-crop experiments on the pepper dataset showed a 2.1% and 2.7% improvement in mAP for object and picking point detection, respectively, compared to the baseline model, confirming its cross-crop robustness. This study reveals the synergistic enhancement of dynamic upsampling and attention mechanisms in agricultural object detection, providing new insights for lightweight model design in complex scenarios.
Why it matches plant phenotyping methods果実と収穫点の画像ベース検出・キーポイント推定モデルを開発し、複数データセットで性能と頑健性を検証しているため、植物形質取得手法が中心である。
abstractthis study proposes MDAD-YOLO (Multi-dimensional Attention and DySample YOLO), a detection model improved based on the YOLOv10n-pose framework.
Tomato industry is one of the important parts of agriculture, and the timely diagnosis of leaf diseases plays a key role in ensuring its production safety. At present, most of the mainstream recognition technologies are based on laboratory standard images to construct recognition models. Although they can achieve accurate analysis of isolated leaves, it is difficult to cope with the practical challenges such as complex climate and environmental factors of plant growth in open environments. The recognition is difficult due to the influence factors of open environment, such as the distortion of disease spots caused by light, the occlusion of branches and leaves, and the confusion of soil attachment and real disease spots. To solve the problem of unbalanced distribution of tomato leaf disease samples in open environment and the big difference of similar diseases, this study proposes a single-modal recognition architecture for tomato leaf diseases based on Efficient localization and Physical information and Dynamic adaptive optimization Network (EPDNet). Firstly, an efficient positioning feature enhancement module is designed to effectively enhance the network’s attention to important regions by calculating and fusing the horizontal and vertical attention weights. Then, a physical information neural network-cross entropy hybrid loss function was designed to ensure the accuracy of prediction, while restricting the smoothness and continuity of the feature map to improve the robustness of the model. Finally, a dynamic adaptive optimization algorithm was designed to iteratively update the learning rate to improve the feature discrimination ability, so as to reduce the identification differences of diseases within the class. Experimental results show that the accuracy of EPDNet on the tomato leaf dataset reaches 93.62%, and the F1 score reaches 93.31 %, which is significantly better than the existing methods. This study provides an effective solution for the application of deep learning methods in crop diseases.
Why it matches plant phenotyping methodsトマト葉の病斑・病害状態を画像から認識する深層学習手法を開発し、データセットで性能評価しているため、植物病害フェノタイピング手法が中心である。
abstractthis study proposes a single-modal recognition architecture for tomato leaf diseases based on Efficient localization and Physical information and Dynamic adaptive optimization Network (EPDNet).
Timely monitoring of grain carbon and nitrogen accumulation dynamics is crucial for the growth monitoring and efficient field management of winter wheat. However, traditional destructive sampling methods are time-consuming, costly, and challenging to implement for large-scale rapid monitoring. Based on the source-sink theory, this study proposes a new method for predicting the dynamic changes of grain carbon and nitrogen accumulation in winter wheat grain by leveraging UAV-based inversion of agronomic parameters (APs). Remotely sensed aboveground biomass and plant nitrogen accumulation were employed as source indicators during the anthesis stage, along with the days after anthesis represented by phenological indices, as co-input variables for the model. Piecewise ordinary least squares regression was employed to analyze the temporal dynamics of grain carbon and nitrogen sink indicators. Combining feature selection and machine learning algorithms, the study developed a UAV multi-spectral image-driven APs inversion framework and visualized relevant grain indicators. The UAV-based grain carbon and nitrogen accumulation prediction model demonstrated excellent performance, with the grain weight accumulation showing R² = 0.86, nRMSE = 21.96 %, and RPD = 2.63; for grain nitrogen accumulation, R² = 0.71, nRMSE = 34.11 %, and RPD = 1.87; and for grain nitrogen content, R² = 0.70, nRMSE = 20.93 %, and RPD = 1.82. The grain carbon and nitrogen accumulation prediction model based on UAV multispectral images enables non-destructive prediction of the entire filling period, and has showcasing high accuracy and application potential.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像から冬コムギ粒の炭素・窒素蓄積などの形質を非破壊推定するモデルと反転フレームワークが研究の中心であり、性能評価も実施している。
abstractthis study proposes a new method for predicting the dynamic changes of grain carbon and nitrogen accumulation in winter wheat grain by leveraging UAV-based inversion of agronomic parameters (APs).
Timely and accurate crop yield estimation is important for sustainable agricultural planning and resource optimization. This study is motivated by the need for a scalable, non-destructive, phenology-aware yield estimation pipeline that can outperform spectral index-based methods. A novel in-season crop yield estimation framework is presented that uses high-resolution UAV-based multispectral imagery and deep neural networks. The pipeline integrates automated phenological stage mapping using a custom Spatial Phenology Attention and Feature Cross (SPARC) Network, canopy structure modeling, and wheat head segmentation via a U-Net model fine-tuned on masks generated with SAM 2. UAV imagery is collected across 18 timestamps, processed to produce reflectance maps, vegetation indices (VIs), canopy height models (CHMs), and fractional cover maps. Plot-level phenological and morphological features are extracted to train multiple regression models for in-season yield estimation. Results show that combining temporal phenological features with structural head metrics significantly improve estimation accuracy, with Gradient Boosting Regression achieving an R2 of 0.89. The proposed approach not only improves the granularity and timeliness of in-season yield estimations but also enables scalable, non-destructive crop monitoring solutions, providing practical information for both farmers and breeders alike.
Why it matches plant phenotyping methodsUAV画像と深層学習を用いて、作物のフェノロジー、形態特徴、穂形状を抽出し、圃場区画レベルの収量を推定する技術パイプラインが研究の中心である。
abstractA novel in-season crop yield estimation framework is presented that uses high-resolution UAV-based multispectral imagery and deep neural networks.
Maize leaf nitrogen exhibits significant vertical heterogeneity within the canopy, which often compromises the accuracy of remote sensing-based nitrogen status diagnosis. To address the limited consideration of leaf-layer contributions and multi-source feature integration, this study proposed a three-dimensional coupled nitrogen diagnosis framework integrating progressive leaf-layer labeling, multi-source feature fusion, and machine learning modeling, based on spring maize experiments under various water and nitrogen regimes in Xinjiang, China. Stratified ground sampling was conducted to obtain leaf nitrogen weight (LNW) from the upper, middle, and lower canopy layers, and 22 vegetation indices (VIs), eight texture features (TFs), and 15 texture indices (TIs) were extracted from UAV multispectral imagery to construct a multi-source feature set. Results demonstrated that the combination of upper and middle canopy leaves best represented overall plant nitrogen status. Feature fusion significantly enhanced model performance, with extreme gradient boosting achieving the highest estimation accuracy for the nitrogen nutrition index (NNI) (R² = 0.68, RPD = 1.77), and convolutional neural networks performing best in LNW estimation (R² = 0.83, RPD = 2.31). The critical nitrogen dilution curves derived from the estimated LNW revealed that irrigation levels significantly influenced the curve intercepts and slopes, highlighting the dual regulatory effects of water on nitrogen uptake and dilution. Coupled with ArcGIS-based spatial visualization, the framework enabled dynamic monitoring of NNI across V6, VT, R3, and R6 growth stages. Overall, this framework effectively improved both the spatiotemporal resolution and estimation accuracy of maize nitrogen status, providing a theoretical basis and technical support for precision nitrogen management and the transition toward sustainable agriculture.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像と多源特徴融合・機械学習により、トウモロコシの葉窒素量および窒素栄養状態を推定する手法を開発・評価しており、フェノタイピング手法が研究の中心である。
abstractthis study proposed a three-dimensional coupled nitrogen diagnosis framework integrating progressive leaf-layer labeling, multi-source feature fusion, and machine learning modeling
Every year, 20%–40% of the global harvest is lost to pests and diseases, underlining the need for rapid and accurate diagnosis. Precision agriculture exploits intelligent devices, such as robots and drones, to enable early detection of pathogens through non-destructive imaging techniques and AI processing. In this study, we exploit Deep Learning techniques for handling multispectral images in agriculture field. In particular, we introduce an adaptive Multi-Model Ensemble framework that processes multispectral data without dimensionality reduction, fully exploiting spectral information to improve early disease detection. Furthermore, several comparisons with dimensionality reduction and data combinations were conducted, exploring different image stack configurations to find the optimal solution in disease detection. We validated our approach on a dataset of tomato plants affected by Tuta Absoluta and Leveillula Taurica, where it improves the ability of disease identification and classification even at early developmental stages, offering promising perspectives for phytosanitary monitoring and sustainable resource management.
Why it matches plant phenotyping methodsトマトの病徴・病害状態をマルチスペクトル画像から識別する深層学習アンサンブル手法を開発し、病害データセットで検証しており、植物フェノタイピング手法が研究の中心である。
abstractwe introduce an adaptive Multi-Model Ensemble framework that processes multispectral data without dimensionality reduction
The grain flow sensor is a core component for achieving precise online yield measurement in combine harvesters. However, the accuracy and stability of sensor monitoring are affected by factors such as the harvester structure, operational vibrations, dust, and electromagnetic interference. Improving sensor performance under these constraints has long been a key research focus in the industry. To enhance the monitoring accuracy and stability of grain mass flow rate in combine harvesters, this study proposes a flow-guided and weighing-based grain mass flow sensor (GMFS), which incorporates flow-guiding, buffering, and flow-stabilising functions. Based on dynamic analysis, a multivariable dynamic monitoring model for the grain mass flow rate was developed, integrating key parameters such as screw conveyor speed and grain weight. Dedicated circuits for weak signal amplification, noise filtering, and signal acquisition were developed, and an improved Kalman filter (KF) was employed to enhance monitoring precision. Building on this foundation, a GMFS prototype was integrated into combine harvesters and tested, and its buffering and deceleration performance was validated through simulations. The simulation results indicated that the average particle velocity magnitude at the GMFS outlet was lower and more stable than at the inlet, confirming the sensor’s effectiveness in buffering and decelerating grain flow. Bench test results showed that, under varying mass flow conditions, the GMFS achieved a signed mean relative error (MRE) of 0.64 % and a root mean square error (RMSE) of 0.041 kg·s⁻¹ for the steady-segment (Δt₁) flow rate, with 95 % confidence intervals (CIs) reported for the MRE. Dynamic tests on a combine harvester demonstrated comparable monitoring accuracy, with an MRE of 0.97 % (95 % CI: 0.31–1.63 %) and an RMSE of 0.107 kg·s⁻¹ for the steady-segment flow rate, and an MRE in full-interval total mass of –0.28 % (mean absolute error: 1.10 %). This study provides an accurate and stable sensor design for real-time monitoring of grain mass flow in combine harvesters and provides robust support for yield monitoring technology.
Why it matches plant phenotyping methodsコンバイン収穫時の穀粒質量流量（収量）をリアルタイム計測するセンサーを設計・実装し、シミュレーション、ベンチ試験、実機試験で精度を検証しており、植物の収量形質取得法が中心的である。
abstractTo enhance the monitoring accuracy and stability of grain mass flow rate in combine harvesters, this study proposes a flow-guided and weighing-based grain mass flow sensor (GMFS)
Field / plotRGB / grayscaleWhole plant / canopy / plot / fieldClassificationSegmentationGrowth / time-series analysisGrowth / development / phenology
Traditional remote sensing approaches for agricultural monitoring are frequently constrained by limited temporal resolution, discontinuous coverage, and high operational costs. These limitations hinder the delivery of high-frequency, continuous, and unmanned agro-information that is critical for precision agriculture—especially in smallholder and autonomous farming scenarios. Meanwhile, communication towers equipped with video cameras are ubiquitously deployed in rural landscapes yet remain underutilized for agricultural applications. This study presents a high-frequency, tower-based, unmanned agricultural monitoring framework that progressively addresses three fundamental challenges, thereby enabling real-time, parcel-level agro-monitoring. In particular, our framework tackles: (1) accurate geo-referencing of oblique imagery through a quaternion-based geographic coordinate transformation, which precisely maps video streams to field parcels; (2) robust segmentation of cultivated parcels via a GIS-guided approach that integrates field boundary data with the Segment Anything Model (SAM) for automatic delineation; and (3) comprehensive temporal intelligence by leveraging a large language model (LLM)-based recognition strategy that fuses time-series imagery, crop rotation history, and field management data to identify crop types, growth stages, farming operations, and anomalies at the parcel level. The resulting hierarchical framework combines front-end local processing for initial event detection with cloud-based advanced analysis, enhancing operational efficiency while ensuring rapid responses. Field deployments in three agricultural regions of China demonstrate that the framework supports automated, hourly-to-daily monitoring of farming activities and crop conditions within a 1–2 km radius of each tower, with key agricultural events, growth stage transitions, and parcel-level anomalies tracked and mapped in near real time. Compared to conventional satellite and aerial remote sensing, this approach offers substantial improvements in monitoring frequency, spatial continuity, and labor efficiency.
Why it matches plant phenotyping methods塔載カメラ画像とAIによって作物の生育段階・異常などの植物状態を時系列推定する監視基盤が研究の中心であり、単なる農業実験のルーチン測定ではない。ただし作業監視や作物識別も含むため、植物表現型への焦点は部分的である。
abstractThis study presents a high-frequency, tower-based, unmanned agricultural monitoring framework
Three-dimensional phenotyping technology is paramount in the field of peanut breeding and cultivation. The intricate topological structure of plants substantially complicates the development of effective peanut phenotyping technologies. In this study, we present the development of a point-cloud-based pipeline for three-dimensional phenotypic analysis of peanut plants. An efficient multi-view image acquisition system and three-dimensional reconstruction techniques were employed to generate point clouds of peanut plants. A dataset comprising 188 labelled samples of peanut point clouds was constructed for the development of semantic and leaf-instance segmentation models based on the transformer architecture. The segmentation accuracy of these models surpassed that of the conventional general segmentation techniques for plant point clouds. Based on the results of the segmentation, 11 three-dimensional phenotypic traits were automatically calculated at both the plant and leaf scales. Among these, five phenotypic traits, including plant height and leaf length, exhibited a mean absolute percentage error (MAPE) of less than 0.12 compared to the measured values. In addition, the Jensen-Shannon divergence (JS divergence) between the probability distributions of the three leaf phenotypic traits and their corresponding measured values was below 0.1. The three-dimensional phenotypic analysis pipeline developed in this study exhibited satisfactory generalisation capabilities, thereby offering an efficacious and expeditious high-throughput phenotyping analysis instrument for the intelligent breeding and cultivation of peanuts.
Why it matches plant phenotyping methodsピーナッツの3D画像取得、点群再構成、分割、形質自動算出を統合したフェノタイピングパイプラインの開発と精度検証が中心である。
abstractwe present the development of a point-cloud-based pipeline for three-dimensional phenotypic analysis of peanut plants
Three-dimensional high-throughput plant phenotyping technology offers an opportunity for simultaneous acquisition of plant organ traits at the scale of plant breeders. Wheat, as a multi-tiller crop with narrow leaves and diverse spikes, poses challenges for organ segmentation and measurement due to issues such as occlusion and adhesion. Therefore, building on previous research, this paper establishes a phenotyping pipeline and develops a 3D phenotypic automated analysis system for individual wheat plants at different growth stages. This system enables automated and precise three-dimensional phenotypic acquisition and analysis of wheat plant architecture, spike morphology, and flag leaf traits. To address the challenges posed by the significant structural differences among wheat spikes, leaves, and stems, as well as their compact spatial distribution, we propose a point cloud segmentation model based on deep learning called ICFMNet. ICFMNet relies on an instance center feature matching module, which extracts features from each instance’s central region and matches them with global point-wise features by computing feature similarity. This approach enables precise instance mask generation independent of the spatial structure of the point cloud. In the analysis of wheat phenotypes, we introduce a contour-based method to accurately extract the barren segment from 3D-scale wheat spikes. Furthermore, we perform the analysis of a total of 19 phenotypes, including flag leaf phenotypes and whole-plant phenotypes. In the organ point cloud segmentation tests for wheat spikes, stems, and leaves, the semantic segmentation achieves mPrec, mRec, and mIoU values of 95.9 %, 96.0 %, and 92.3 %, respectively. The instance segmentation attains mAP and mAR scores of 81.7 % and 83.0 %, respectively. Moreover, in comparison to five other segmentation network models, ICFMNet demonstrates superior segmentation performance. To better assess barren segment localization accuracy, additional evaluations are conducted using two metrics: interval overlap and interval error, achieving values of 92.33 % and 0.1123 cm, respectively. Experimental results indicate that our method excels in terms of accuracy, efficiency, and robustness, providing a reliable systematic platform for precise identification and breeding research of wheat plant types. The source code and trained models for ICFMNet are available at https://github.com/xiao-pl/ICFMNet.
Why it matches plant phenotyping methods小麦個体・器官の3D形質を自動取得・抽出するセグメンテーションおよび解析パイプラインの開発と技術評価が研究の中心である。
abstractthis paper establishes a phenotyping pipeline and develops a 3D phenotypic automated analysis system for individual wheat plants at different growth stages.
Under the intensifying impact of global climate change, the frequency of plant disease outbreaks is steadily increasing, posing significant challenges to healthy cultivation and precision management. To enable early diagnosis of plant disease stress, this study used two rose (Rosa chinensis) varieties, ’Red Cap’ and ’Carefree Wonder’ were used to conduct powdery mildew stress experiment. Throughout the stress period, leaf electrical impedance spectroscopy (EIS), physiological parameters, and ultrastructural observations, were collected. A novel lumped equivalent circuit model was proposed, incorporating plant cell electrophysiological properties. The model developed for both rose varieties-featuring Constant Phase Elements (CPE) and Warburg elements (W), successfully characterized the resistive properties of the leaf tissue, including the extracellular resistance (R₁), cell membrane resistance (R₂), intracellular resistance (R₃), and vacuole interior resistance (R₄). Model parameters R₁ and R₃ were significantly correlated with the above physiological indicators, and showed significant differences 3 to 11 days earlier than traditional physiological parameters, demonstrating strong potential for early detection of cellular damage. Overall, this study demonstrates that EIS technology can dynamically reflect electrical property changes in plant tissues under biotic stress, effectively overcoming the lag limitations of conventional physiological measurements, providing a promising tool for early disease diagnosis and resistance screening.
Why it matches plant phenotyping methodsバラ葉の病害ストレス状態を電気インピーダンス分光法で測定し、等価回路モデルを開発・検証して早期診断性能を評価しているため、植物フェノタイピング手法が中心である。
abstractA novel lumped equivalent circuit model was proposed, incorporating plant cell electrophysiological properties.
Cost-effective remote sensing solutions are critically needed to democratize precision agriculture technologies. While hyperspectral and LiDAR systems deliver high accuracy, their prohibitive costs limit widespread adoption. This study demonstrates that systematic multi-modal feature integration transforms standard UAV-based RGB imagery into a powerful phenotyping instrument, achieving crop trait prediction accuracy comparable to systems costing 10–50 times more. We developed a comprehensive framework integrating spectral indices, geometric parameters, and texture metrics from commodity RGB sensors to predict five critical cotton traits: leaf area index (LAI), intercepted photosynthetically active radiation (IPAR), above-ground biomass, lint yield, and seed cotton yield. The progressive integration approach employed Random Forest regression with four feature configurations: baseline color indices (CIbₐₛₑ), refined color indices (CIᵣₑf), geometric parameters (CIᵣₑf + GP), and texture metrics (CIᵣₑf + GP + T). Field experiments across three trials over two growing seasons (2022–2023) with varying genotypes, planting densities, and sowing dates provided 2,126 ground truth measurements for model development and validation. The optimal multi-modal model achieved R² = 0.97 for IPAR (rRMSE = 6 %), R² = 0.91 for LAI (rRMSE = 15 %), and R² = 0.85 for biomass (rRMSE = 32 %), with lint yield and seed cotton yield demonstrating R² values of 0.92 and 0.77, respectively. Variance partitioning analysis revealed texture features as the dominant contributor (16.2 % ± 7.1 %), followed by spectral indices (9.1 % ± 4.2 %) and geometric parameters (8.0 % ± 2.8 %), with substantial shared variance (45–65 %) indicating strong feature complementarity. Phenological analysis demonstrated that flowering-stage imagery outperformed boll opening stage measurements, while stage-general models showed superior robustness. Cross-temporal validation confirmed model generalizability, with trial-general models achieving R² values of 0.91–0.97 for IPAR across diverse environmental conditions. The framework enables sub-meter spatial resolution trait mapping while maintaining operational simplicity and cost-effectiveness, demonstrating that systematic feature engineering can democratize high-precision phenotyping technologies for broader agricultural applications.
Why it matches plant phenotyping methodsUAV-RGB画像から複数の綿形質を推定する特徴統合フレームワークを開発・検証しており、形質取得・抽出手法が研究の中心である。
abstractThis study demonstrates that systematic multi-modal feature integration transforms standard UAV-based RGB imagery into a powerful phenotyping instrument
Vertical indoor hydroponic farms offer sustainable solutions in land scarce countries to foster agriculture productivity for addressing growing demand. Such farms require extensive controllability of the growing conditions to ensure year round-cultivation of diverse crops within the space available. Continuous monitoring of the crops and early remedial measures are essential to ensure non-compromised, high-quality yield from these farms. Currently, most farms rely on human vision based monitoring, which is quite subjective and time-consuming and could be ineffective in identifying crop stresses at early stages. Hence, efficient management of these farms requires advanced automated systems to monitor crop health, including possible stress factors such as nutrient, water, and light deficiencies at early stages to enable timely intervention. This research, in this context, explores innovative strategies using assessment parameters such as spectral ratios and derivative reflectance derived from hyperspectral images for crop monitoring. Customized spectral index for nutrient deficiency detection and approaches for quantification of derivative spectra for stress detection are developed. These strategies can be used to rapidly detect the stresses at the early stages non-destructively (within hours in case of light and water deficiencies) and could promptly guide in timely remedial actions. The proposed method offers automation possibilities for non-invasive monitoring systems utilizing hyperspectral vision. This non-invasive imaging system integrated on a robotic platform is envisaged to revolutionize the development of unmanned indoor hydroponic farms for a sustainable future.
Why it matches plant phenotyping methodsハイパースペクトル画像から栄養・水・光ストレスを早期検出・定量する手法を開発しており、植物状態の取得方法が研究の中心である。
abstractCustomized spectral index for nutrient deficiency detection and approaches for quantification of derivative spectra for stress detection are developed.
As an important economic crop in tropical regions, the natural rubber yield of rubber trees is closely related to their crown structure. Accurately extracting tree crowns is fundamental for obtaining key growth parameters and evaluating yield potential. However, existing methods face three major challenges when processing LiDAR point cloud data of rubber trees: ambiguous boundaries due to complex canopy structures, difficult segmentation caused by background interference, and learning rate optimization issues. To address these challenges, this paper proposes a single-tree crown extraction method based on UAV LiDAR point clouds (RTCrownNet). First, a Dual-Stream Collaborative Feature Fusion Module (DS-CFM) is designed to integrate local geometric details and global semantic information, enabling accurate identification of complex crown boundaries. Second, a Residual-Augmented Graph Convolution Module (RAGC) is proposed to encode the topological relationships of point clouds using graph structures, enhancing the model’s ability to distinguish between overlapping leaves and ground areas. Additionally, an Adaptive Coati Differential Evolution Algorithm (ACDE) is developed, which constructs a dual-track parallel search framework to automatically optimize learning rates, accelerate model convergence, and enhance generalization performance. Experimental results show that RTCrownNet outperforms three traditional methods and seven deep learning networks on a self-built rubber tree point cloud dataset, achieving an instance mean intersection over union (mIoU) of 87.31% and an F-score of 95.24%. In generalization experiments, the method demonstrates excellent performance on the Wytham Woods temperate deciduous forest dataset and the FOR-instance dataset covering different forest types in five countries, verifying the model’s versatility. This study provides reliable technical support for precise monitoring, intelligent management, and resource evaluation of rubber trees, and holds significant importance for promoting the sustainable development of the rubber industry.
Why it matches plant phenotyping methodsUAV LiDAR点群からゴム樹の単木樹冠を抽出する手法を開発・比較検証しており、樹冠構造という植物形態形質の取得が研究の中心である。
abstractAccurately extracting tree crowns is fundamental for obtaining key growth parameters and evaluating yield potential.
Leaf protein content (LPC) is a critical physiological parameter for assessing crop nitrogen status, optimizing fertilization strategies, and predicting crop yield. Although hyperspectral remote sensing offers a nondestructive alternative, it still faces challenges such as overlapping protein and water absorption spectral features. This study presents a “physics-constrained + data-driven” hybrid modeling framework, LPCNet, for remote sensing–based LPC estimation. The core innovations of LPCNet include (1) leveraging the physics-based PROSPECT-PRO and SAIL radiative transfer models to generate a simulated spectra dataset, addressing the challenges of small sample sizes and distribution bias in field measurements through pretraining and transfer learning; (2) incorporating leaf chlorophyll content (LCC) as an auxiliary training target within a multitask learning framework, which exploits the strong absorption features of LCC in the visible–near infrared (VNIR) range to enhance the ability of the model to interpret weak LPC absorption signals in the shortwave infrared (SWIR) range; and (3) employing a multiscale convolutional network with a feature fusion mechanism to explicitly model the complex nonlinear relationships between spectral reflectance and LPC. This study utilized field-measured data from three growing seasons of wheat and potato to develop and validate the LPCNet model. The results demonstrate the following: (1) the LPCNet model pretrained with a simulated spectra dataset notably outperforms nonpretrained models; (2) the pretrained and LCC-assisted strategy further improves LPC-estimation accuracy to RMSE = 0.000100 g/cm² (R² = 0.866), showing a substantial advantage over traditional RF (R² = 0.760, RMSE = 0.000134 g/cm²). This study proposes a hybrid deep learning modeling framework utilizing hyperspectral remote sensing for high-precision monitoring of crop LPC.
Why it matches plant phenotyping methodsハイパースペクトル計測から作物葉のタンパク質含量という植物生理形質を推定するモデルを開発し、複数年の圃場データで検証しており、表現型取得・推定手法が中心である。
abstractThis study presents a “physics-constrained + data-driven” hybrid modeling framework, LPCNet, for remote sensing–based LPC estimation.
Accurate maize tassel counting on individual ridge plays a critical role in advancing maize breeding programs by providing insights into crop adaptability and agronomic performance. The current manual method of processing male inflorescences has low accuracy and high labor intensity. Moreover, the existing algorithms cannot be directly applied to the counting of an individual ridge maize tassels. An automated system with UAV-based RGB imagery is presented for maize tassel counting. The object detection model was developed based on You Only Look Once version 8-medium (YOLOv8m), which detects tassels. A double-step Otsu threshold algorithm (DSOTSUTA) was designed to extract individual maize tassel ridge, which eliminated the interference of two adjacent ridges on both sides. Individual ridge tassel counting was implemented by Deep learning based Simple Online and Realtime Tracking (Deep SORT). It assigned identifiers (IDs) to each tassel and filtered abnormal IDs by analyzing the displacement increments of IDs in consecutive frames eliminating errors caused by ID switching. The object detection model achieved a mean Average Precision (mAP) of 91.6 %. The DSOTSUTA was tested on 5340 images and effectively extracted individual maize tassel ridges. The system achieved a root mean square error (RMSE) of 22.14 tassels per video, a mean absolute percentage error (MAPE) of 8.43 %, and an accuracy of 92.23 %, signifying a mean absolute percentage error (MAPE) of 8.43 % between the predicted tassel counts and ground truth observations. These results indicate that this automated system has the ability to enhance the accuracy of individual ridge maize tassel counting in breeding programs.
Why it matches plant phenotyping methodsUAV画像からトウモロコシ雄穂を自動検出・追跡・計数する手法を開発し、精度評価まで行っており、植物形質取得が研究の中心である。
abstractAn automated system with UAV-based RGB imagery is presented for maize tassel counting.
Existing tasks such as fruit growth monitoring, harvesting, and quality sorting still suffer from low precision and insufficient automation. 3D reconstruction technology can accurately capture the external characteristics of fruits and shows great potential for enhancing automated fruit detection and processing. This paper, guided by two core application needs in the fruit industry: real-time online sensing and offline high-precision analysis, provides a detailed overview of research progress in 3D reconstruction technology for fruits. It first introduces the principles, workflows, and advantages and limitations of classical 3D reconstruction methods. Then, it focuses on the basic framework of learning-based 3D reconstruction approaches, their improvement directions, and their applications in fruits and other agricultural products. In addition, the challenges encountered in fruit 3D reconstruction, such as occlusion and complex lighting conditions, are summarized, along with potential solutions. Finally, future research directions are discussed. This review serves as a valuable reference for promoting the integration of computer vision and agricultural intelligence and advancing the fruit industry chain’s digital and intelligent transformation.
Why it matches plant phenotyping methods果実の外部形質を取得する3D再構成手法を中心に、原理・ワークフロー・限界・応用・課題を体系的にレビューしており、植物フェノタイピング手法のレビューに該当する。
titleProcess, challenges and solutions of fruit 3D reconstruction: A review
Maize (Zea mays L.) is a crucial grain and economic crop with extensive applications in food, feed, and industry. Phenotypic traits such as stem circumference (SC), stem height (SH), and the stem circumference-to-height ratio (SCHR) are essential indicators for studying maize development, environmental adaptation, and lodging resistance. Traditional manual measurement methods are inefficient, costly, and unsuitable for large-scale phenotypic monitoring. While unmanned aerial vehicle (UAV)-based approaches have achieved relatively accurate SH estimation, SC estimation remains challenging using UAV technology alone, limiting SCHR estimation. This study combined digital camera sensors on unmanned ground vehicle (UGV) and UAV platforms to capture maize stem and canopy images, enabling the estimation of SC, SH, and SCHR. The primary contributions of this study are as follows: (1) We propose the maize-stem segmentation network for calculating stem diameter and circumference (MSSDCNet) to segment maize stems in images and estimate SC based on the segmentation results. (2) We process UAV-derived digital surface models to extract SH information and employ a linear regression (LR) model for SH estimation. (3) Using the estimated SC and SH, we calculate SCHR and analyze its temporal variations across different growth stages. The results demonstrate that: (1) MSSDCNet accurately segments maize stems from images and facilitates SC estimation (R² = 0.759, RMSE = 0.414 cm, nRMSE = 0.087). Temporal analysis of SC reveals a gradual decrease during the reproductive growth stage, potentially due to the transfer of photosynthetic products to the maize cob and stem water loss. (2) This study accurately estimated SH (R² = 0.941, RMSE = 0.151 m, nRMSE = 0.078). However, SH estimates during the reproductive growth stage tend to be underestimated, likely due to DSM point clouds being more sensitive to sharp features such as tassels. (3) SCHR estimation achieves R² = 0.453, RMSE = 2.287 × 10⁻³, nRMSE = 0.136. Temporal analysis reveals a general decline in SCHR from the kernel blister stage to the dough stage, with some maize materials consistently exhibiting lower SCHR levels during each growth stage. This may be related to genetic traits, planting density, soil fertility, or other environmental factors. By integrating UGV-based maize stem images and UAV-based maize canopy images with MSSDCNet and LR, this study successfully estimates SC, SH, and SCHR for various maize materials. This study provides a novel technique for maize, early lodging risk prediction, and lodging-resistant breeding lines screening, contributing to rapid and efficient maize phenotypic monitoring under field conditions.
Why it matches plant phenotyping methodsUAV・UGV画像、DSM、深層学習によってトウモロコシの茎周囲長・草丈・比率を推定する手法を開発・評価しており、植物表現型取得が研究の中心です。
abstractThis study combined digital camera sensors on unmanned ground vehicle (UGV) and UAV platforms to capture maize stem and canopy images, enabling the estimation of SC, SH, and SCHR.
The goal of automated crop leaf disease detection (CLDD) is to extract fine-grain visual features from images for classification. Vision transformers (ViTs) and attention-based models have improved classification accuracies by narrowing down the receptive fields of visual features, though often at the expense of computations and robustness. Generalizing visual features becomes difficult due to lacking diversification and overrepresentation of healthy leaves over diseased ones, causing overfitting and inherently limiting ViT’s capacity. Vision Transformers (ViTs) typically rely on single feature projection onto query, key, and value embeddings for self-attention calculations. In contrast, we introduce multi-feature projection. The texture representations extracted from RGB images are fed into key and value components, while the query takes RGB-coded features. Attention is the SoftMax-activated dot product computed on the three components. Simultaneously, a residual branch integrates RGB features with the attention vectors, producing moderately robust and interpretable compositions referred to as Texture Guided Visual Attention (TGVA) features. These TGVA features are integrated back into the backbone classifier network. Evaluating the Texture Guided Visual Attention neural network (TGVAnn) on the most challenging plant leaf datasets, sugarcane and maize, demonstrates its superior performance, achieving a 9% improvement in classification accuracy over traditional ViT models. Furthermore, TGVAnn shows a 20%–70% reduction in theoretical computational requirements (GFLOPs) relative to comparable baselines under a common reporting convention, supporting efficiency and scalability. This method outperforms similar approaches in both accuracy and robustness. In conclusion, the TGVA features derived from the multi-feature attention module effectively enhance the robustness of backbone image classifiers. This improvement is achieved with only a minimal increase in computational overhead of 13 M-parameters, making the approach both efficient and suitable for edge deployment scenarios.
Why it matches plant phenotyping methods植物葉画像から病害状態を推定する画像解析手法を開発し、複数データセットで性能評価しており、フェノタイピング手法が中心である。
abstractThe goal of automated crop leaf disease detection (CLDD) is to extract fine-grain visual features from images for classification.
Accurately estimating individual plant evapotranspiration is essential for precise management and sustainable resource use in greenhouse cultivation. Integrating evapotranspiration models with crop-monitoring devices capable of acquiring images and solar radiation data may enable plant-level estimation of crop evapotranspiration. In this study, a plant-specific crop evapotranspiration estimation system was developed for hydroponic tomato cultivation in greenhouses during the harvest season. The evapotranspiration was estimated using a simplified Penman–Monteith model based on the leaf area index (LAI), solar radiation, air temperature, and relative humidity. The model was subsequently generalized through z-score normalization. To acquire side-view RGB images of individual tomato plants and measure the solar radiation distribution, a rail-based crop-monitoring device was employed. A ResNet-based convolutional neural network model was developed to estimate the LAI from the acquired images. The images were augmented via permutations with repetition to enhance the model’s accuracy. An image-merging method and a You Only Look Once version 8 Nano-based object detection model were used for rapid and automated image acquisition. The system calculated the crop evapotranspiration for each plant, and its performance was evaluated in a tomato cultivation greenhouse. Validation tests revealed strong correlations between the estimated and measured LAI (R² = 0.89, RMSE = 0.06) and between the predicted and actual evapotranspiration values (R² = 0.88, RMSE = 26.43 g h⁻¹ plant⁻¹). Distribution maps for the LAI and evapotranspiration were generated using the developed system. The system can accurately assess plant-specific evapotranspiration, thereby supporting precision crop management and helping improve productivity in greenhouse cultivation.
Why it matches plant phenotyping methods個体別のLAI画像推定と蒸発散量推定を中核とする監視システムを開発し、実測値との相関で検証しているため、植物表現型取得・推定手法として対象に含める。
abstracta plant-specific crop evapotranspiration estimation system was developed for hydroponic tomato cultivation in greenhouses
Multispectral image analysis is an effective way to detect crop growth status. However, the complexity of manufacturing process and technology of multispectral image acquisition equipment make data acquisition expensive. Therefore, a method based on a window-adaptive spatial-spectral attention transformer is proposed to reconstruct multispectral images using RGB images of maize. First, RGB and hyperspectral images of the maize are obtained, and the reflectance data from classic and preferred band combinations are extracted from the hyperspectral image. Then, a transformer model is constructed to evaluate and compare the reconstruction efficacy of the 5-band and 10-band combinations across four attention modes: spatial, spectral, spatial-spectral, and window-adaptive spatial-spectral attention. The best-performing reconstruction results are selected and compared with the original data from three perspectives: image, spectrum, and model effect. The 10-band multispectral image reconstructed by the window-adaptive spatial-spectral attention mechanism is highly similar to the original image, with a reflectance correlation exceeding 0.99. Furthermore, its application in monitoring crop growth status (i.e., maize chlorophyll) yields results closely aligned with actual reflectance data: RC² is 0.76, RV² is 0.64, while RMSEC and RMSEV are 3.63 mg/L and 2.94 mg/L, respectively. To further explore the model performance, the new sensitive bands are selected to be reconstructed in the maize V7 stage. The results from the chlorophyll content prediction model are as: RC² is 0.64, RV² is 0.60, with RMSEC and RMSEV are 5.61 mg/L and 5.62 mg/L, respectively. Therefore, the window-adaptive spatial-spectral attention transformer can accurately reconstruct multispectral images and establish precise growth status monitoring models, providing technical support for low-cost field maize growth detection.
Why it matches plant phenotyping methodsRGB画像からマルチスペクトル画像を再構成する手法を開発・評価し、トウモロコシのクロロフィル量（生育状態）推定に適用しているため、植物表現型取得・推定法が中心である。
abstracta method based on a window-adaptive spatial-spectral attention transformer is proposed to reconstruct multispectral images using RGB images of maize.
Plant density is an important variable for management and phenotyping of small-grain cereal crops such as wheat and barley. While many image-based estimation methods exist to replace laborious manual counting, most of them rely on empirical relationships that may not generalize well to different sites, growth stages, species and varieties. In this study, we propose a novel small-grain cereal plant density estimation method that uses leaf tip density dynamics derived from submillimeter-scale images acquired at 45° view zenith angle. This method contained two steps. In the first step, a P2PNet deep learning detection model was trained to estimate leaf tip count in a surface of known area to get the leaf tip density. An occlusion correction method was then applied on this density, leading to an estimation error of about 20% at critical growth stages. In the second step, a wheat leaf dynamic model was used to simulate the evolution of leaf tip density over thermal time as functions of several variables, including mean time of plant emergence, phyllochron and plant density. This model was then inverted using a lookup table approach to estimate plant density from leaf tip density dynamics. The results obtained on three test datasets indicated that two observations performed before the appearance of the second and third leaves could be sufficient to attain a relative plant density estimation error of about 10%. We also discussed that this method should be able to work on other datasets without recalibration, and estimate other variables such as phyllochron at early growth stages. The code will be available at: https://github.com/wdwzytc/WheatPlantDensity.
Why it matches plant phenotyping methodsRGB画像から葉先密度を抽出し、植物密度を推定する画像ベースの表現型計測法を開発・評価しており、方法が研究の中心である。
abstractwe propose a novel small-grain cereal plant density estimation method that uses leaf tip density dynamics derived from submillimeter-scale images
SoybeanLiDAR / point cloudRoot2D/3D reconstructionSegmentationSkeletonization / topologyRoot system architecture
Characterizing root system architecture (RSA) is essential for understanding plant acclimatization and guiding breeding strategies to enhance stress tolerance and optimize resource uptake. Although 3D root analysis provides significantly more detailed and structurally informative insights than conventional 2D methods, the development of robust and quantitative tools for 3D root phenotyping has been hindered by challenges such as data complexity, noise, and root overlap. In this study, we present a biologically inspired skeletonization framework that segments root architectures by tracing root growth trajectories. The primary objective is to enable anatomically accurate extraction of RSA traits from 3D point clouds. Our method begins by segmenting the primary root through shortest-path extraction and tangent-plane-based clustering. Lateral root initiation points are then detected, and candidate paths are grown using a bionic pathfinding strategy with adaptive parameters; an optimal, non-overlapping skeleton is selected through clustering and combination sorting, and finally refined via an inward back-tracing procedure to improve junction connectivity. To support downstream phenotyping, we compute root length and angle from the segmented skeletons, and reconstruct anatomically faithful tubular meshes for each lateral root to analytically estimate surface area and volume. Our method achieved high accuracy across multiple traits, including an F1 score of 0.88 for lateral root numeration, R2 values of 0.992 and 0.987 for primary and lateral root length estimation, respectively, and strong agreement in surface area (R2=0.953) and volume (R2=0.912) validation against reference methods. Overall, our method offers a robust and biologically meaningful solution for 3D root phenotyping. The extracted traits provide plant breeders with critical insights for genotype selection and offer plant scientists a powerful tool to evaluate the effects of agronomic treatments and environmental interventions.
Why it matches plant phenotyping methods3D根系骨架化と形態形質抽出法の開発・検証が研究の中心であり、根長・角度・表面積・体積などの表現型を定量化している。
abstractwe present a biologically inspired skeletonization framework that segments root architectures by tracing root growth trajectories
Under the dual pressures of food security and sustainable agricultural development, rapid and simultaneous detection of multiple crop growth parameters has become a core technological requirement for optimizing field management and improving resource utilization efficiency. UAVs carrying one or more sensors to collect of different crop growth parameters have achieved remarkable results in the field of single morphological or physiological parameter analysis. However, existing low-cost devices often failed to collect 3D geometric data and high-resolution spectral information simultaneously in field conditions, while the different nature and data structure of point cloud and spectral data brought special challenges to data fusion, restricting the ability of simultaneous multi-parameter resolution. Facing such challenges, in this paper, we design and develop a system that can take into account the simultaneous acquisition of 3D geometric data and high-resolution spectral information in field conditions through multi-sensor fusion and deep learning algorithm innovation. Based on a mature color point cloud data structure, we combine RGB cameras and laser radar sensors to fuse RGB images and point cloud data. We improved a spectral reconstruction network, construct a dedicated chlorophyll response sensitive band dataset for training, and reconstructed hyperspectral images with 36 channels in the 500–850 nm band range from RGB images, which greatly reduces the cost of the spectral information acquisition device. Experiments show that the SAM (Spectral Angle Mapper) value between the reconstructed hyperspectral data and the original hyperspectral data is less than 0.03. Finally, the growth parameters of crops are estimated using spectral and point cloud data. The developed equipment was calibrated and tested, and experimental data were collected under real field conditions for plant height (PH), leaf area index (LAI), and chlorophyll content estimation. The experimental results showed that the system could accurately analyze maize PH and LAI with Rt2 of 0.98 and 0.97, respectively, and that the chlorophyll content analysis capability was at the same level as that of other studies that have used UAV-mounted hyperspectral cameras for leaf chlorophyll content (LCC) detection, and the established estimation model Rt2 reached 0.66. The canopy chlorophyll content (CCC) of maize could be accurately estimated by fusing the data, and the Rt2 reached 0.95.
Why it matches plant phenotyping methodsRGB画像・LiDAR融合と深層学習による3D形状およびハイパースペクトル情報の再構成システムを開発し、圃場で植物形質推定を校正・検証しており、フェノタイピング手法が中心である。
abstractin this paper, we design and develop a system that can take into account the simultaneous acquisition of 3D geometric data and high-resolution spectral information in field conditions through multi-sensor fusion and deep learning algorithm innovation.
Rice is a vital staple food for global food security and a primary income source for millions of farmers worldwide. However, abnormal rice growth poses a serious threat to both yield stability and grain quality, undermining agricultural productivity. Early detection of such anomalies is therefore essential to mitigate yield losses. However, existing methods either targeted only one symptom at a time, or failed to generalize under various field conditions. Moreover, lightweight real-time inference is needed for on-board UAV deployment, yet most high-accuracy models incur prohibitive computational cost. In this study, we propose ARG-TR model, a lightweight transformer-based semantic segmentation framework built on the SegFormer architecture, which utilizes long-range dependencies to identify complex growth anomalies. The model is trained and validated on a large-scale, drone-captured multi-spectral dataset. By integrating a hierarchical transformer encoder with a lightweight decoder, ARG-TR achieves rapid convergence during training and demonstrates strong generalization to unseen data. The experimental results on a challenging dataset of abnormal rice growth patterns show that ARG-TR achieves a robust Intersection over Union (IoU) of 64.8, which outperforms state-of-the-art baselines such as MaskFormer and KNet in both accuracy and computational efficiency.
Why it matches plant phenotyping methodsドローンマルチスペクトル画像からイネの異常生育状態を抽出するセマンティックセグメンテーション手法を開発・検証しており、植物状態の取得方法が中心である。
abstractwe propose ARG-TR model, a lightweight transformer-based semantic segmentation framework built on the SegFormer architecture, which utilizes long-range dependencies to identify complex growth anomalies.
Spatial frequency domain imaging (SFDI) is a non-invasive optical imaging technique widely used for the quantitative determination of fruit tissue optical properties, specifically absorption coefficient (μₐ) and reduced scattering coefficient (μₛ’). However, traditional SFDI methods rely on multiple frequency and phase images, limiting real-time imaging capabilities. To address this issue, we present a novel rapid prediction method based on a two-stage deep neural network architecture, termed FSGOP (Frequency-Spatial Attention UNet and GAN-based two-stage network for optical properties prediction). Compared with conventional three-phase demodulation SFDI, this method reduces the acquisition time by approximately 5/6 and requires only 0.21 s for inference. In the first stage, a UNet network enhanced by Frequency-Spatial Attention (FSA) is employed to effectively decouple the multi-frequency components. In the second stage, a Generative Adversarial Network (GAN) is utilized to predict the optical properties, thereby enabling the simultaneous extraction of μₐ and μₛ’ maps under different frequency conditions from a single multi-frequency mixed fringe image. In experiments on apples, pears, and peaches, the method yielded normalized mean absolute errors of 0.10 (f₁) and 0.09 (f₂) for μₛ’, and 0.07 and 0.06 for μₐ, respectively. The results revealed significant complementary information in the optical property maps at different frequencies, with lower frequencies being more sensitive to subsurface damage and higher frequencies revealing surface texture features more effectively. This method enhances information utilization and real-time performance in multi-frequency imaging, offering a rapid, accurate, and low-cost solution for optical property extraction and quality inspection of agricultural products.
Why it matches plant phenotyping methods果実の光学特性を単一画像から推定する画像・深層学習手法を開発し、取得時間と精度を評価しており、植物器官の状態計測が中心である。
abstractwe present a novel rapid prediction method based on a two-stage deep neural network architecture, termed FSGOP
Fruit diameter grading is essential for commercialization, packaging, and market sales, directly impacting the value and competitiveness of fruits. Traditional diameter grading is typically performed after harvesting, relying on manual or mechanical methods. This process introduces extra steps and increases the risk of fruit damage during transit, which can reduce economic efficiency. To address this issue, this study introduces a real-time apple diameter grading method utilizing a flexible force-sensing gripper and the CNN-BiLSTM-Attention deep learning network, enabling synchronized intelligent identification and grading of apple diameter during robotic mechanical picking. First, ionogel-based triboelectric nanogenerators (IG-TENG) were developed and mounted on the surface of a three-finger Fin-Ray flexible picking end effector. A contact force monitoring system was established using a modular apple model, and the force sensor was calibrated. This setup allowed for accurate measurement of the contact force between the fingers and the apple. Using a multi-layer perceptron (MLP) to integrate mechanical response data, robotic hand motor stroke, and apple posture information, an apple contact force model was created to accurately predict the actual gripping force under various grasping conditions. Finally, a CNN-BiLSTM-Attention diameter prediction model was designed to deliver real-time, precise fruit diameter estimates. Orchard experiments demonstrated that the apple diameter grading method, combining force sensing with deep learning, achieved a mean absolute error (MAE) of 2.13 mm and a grading accuracy of 92 %, supporting non-destructive gripping and accurate grading. This research addresses the limitations of traditional diameter grading methods, streamlines harvesting steps, enhances efficiency, and offers a cost-effective and reliable solution for non-destructive fruit diameter grading.
Why it matches plant phenotyping methodsリンゴ径という植物器官形質を、力覚センサーと深層学習でリアルタイム推定・等級化する手法の開発、校正、検証が研究の中心である。
abstractFinally, a CNN-BiLSTM-Attention diameter prediction model was designed to deliver real-time, precise fruit diameter estimates.
As an economically important crop, tobacco requires the precise extraction of phenotypic characterization data, which is crucial for breeding, cultivation practices, physiological research, and industrial applications. However, there is currently a lack of automated algorithms for extracting key basic phenotypic traits such as plant height, leaf number, leaf area, and stem-leaf angle. In this study, we developed a set of computational methods for fully automated extraction of these phenotypic features from 3D tobacco point cloud data. Specifically, our pipeline includes: (1) preprocessing the 3D point cloud data, involving operations such as downsampling, denoising, normal vector estimation, and coordinate transformation; (2) integrating a graph neural network with a region-growing algorithm to segment leaves, stems, and other organs, and refining the segmentation results to address the challenge of overlapping leaves; and (3) calculating fundamental phenotypic attributes including plant height, leaf count, leaf area, and stem-leaf angle based on the segmentation output. Additionally, to address potential gaps in the scanned point cloud, we implemented perforation detection and repair operations. The effectiveness and accuracy of the proposed algorithm were validated through mathematical model simulations. Distinct from traditional statistical discriminative methods, this approach provides a novel framework for the precise extraction of tobacco phenotypic data.
Why it matches plant phenotyping methods3D点群から植物器官を分割し、草丈・葉数・葉面積・茎葉角を自動抽出する計算手法の開発と検証が中心である。
abstractwe developed a set of computational methods for fully automated extraction of these phenotypic features from 3D tobacco point cloud data.
Seed viability is crucial for ensuring crop quality and yield. However, existing nondestructive detection methods, which primarily rely on spectroscopic techniques and simple data fusion strategies, often suffer from limited accuracy and reliability. To address these limitations, this study proposes a novel, highly accurate, and interpretable nondestructive approach for evaluating maize seed viability. With regard to enhancing the prediction accuracy of seed viability, a grouped hyperspectral image fusion (GHIF) strategy was proposed to more effectively integrate complementary information from visible-near-infrared hyperspectral imaging (VisNIR-HSI) and fluorescence hyperspectral imaging (Fluo-HSI) datasets. With respect to improving model interpretability, eight biochemical components in the embryo of maize seeds were measured, and two key biochemical indicators—catalase (CAT) activity and malondialdehyde (MDA) content—were identified and validated as highly correlated with seed viability and predictable from spectral data. Building on these findings, a two-stage detection model was constructed. In the first stage, the two key biochemical indicators were predicted from the fused data using regression models. In the second stage, seed viability was determined using a dual-threshold strategy based on the predicted biochemical values. Experimental results showed that the proposed method achieved 90 % classification accuracy, comparable to direct spectral models while offering greater interpretability. This approach provides a reliable and explainable solution for nondestructive seed viability evaluation.
Why it matches plant phenotyping methodsトウモロコシ種子の生存性という植物状態を、可視近赤外・蛍光ハイパースペクトル画像の融合と解釈可能な予測モデルで非破壊推定する手法が研究の中心である。
abstractthis study proposes a novel, highly accurate, and interpretable nondestructive approach for evaluating maize seed viability
Rice is a fundamental staple crop germplasm and a vital resource for germplasm innovation, playing a critical role in global food security. Viability is a key indicator for evaluating the conservation and utilization of germplasm resources, ensuring high and stable grain yields. Viability loss during the germplasm conservation process is a natural-aging process. Given the large number of varieties and the rarity of certain samples, excessive destructive tests for viability assessment should be minimized and ultimately replaced by intelligent non-destructive detection methods. Therefore, it is imperative to explore intelligent non-destructive, few-shot, cross-variety/germplasm, and viability detection of rice germplasm based on natural-aging. Current algorithms for rice germplasm viability detection encounter significant challenges in achieving optimal performance under few-shot conditions. We propose a spectral Kolmogorov-Arnold Transformer algorithm, specifically designed for viability detection of rice germplasm under few-shot conditions. A feature enhancement module is implemented to improve the spectral feature representation capabilities of germplasm hyperspectral image (GHSI). A multi-scale spectral feature extraction module is designed to extract spectral features across multiple scales. A fusion of convolutional neural network and Transformer module is introduced to capture both global and local features of GHSI. Finally, a learnable activation function (Kolmogorov-Arnold networks, KAN) and global average pooling are employed for viability classification. Under the condition of 15 samples per class, the SKA-T achieved overall accuracies of 92.87%, 92.30%, and 92.57% for the three rice lines, respectively. These results demonstrate the effectiveness of SKA-T in intelligent non-destructive viability detection of rice germplasm under few-shot conditions.
Why it matches plant phenotyping methodsイネ種子の生存性という植物状態を、ハイパースペクトル画像から非破壊推定する新規アルゴリズムを開発し、複数系統・少数サンプル条件で性能評価しているため、フェノタイピング手法が中心である。
abstractWe propose a spectral Kolmogorov-Arnold Transformer algorithm, specifically designed for viability detection of rice germplasm under few-shot conditions.
Rice leaf blast significantly threatens rice quality and yield, necessitating efficient and precise identification methods for effective field management. Current methods face challenges in accurately detecting leaf blast and distinguishing dense targets due to their small size, scale variation, and dense distribution. This paper proposes a lightweight rotational rice leaf blast detection algorithm named Ro-YOLOv8-PKI. The algorithm adopts Oriented Bounding Boxes (OBB) over traditional Horizontal Bounding Boxes (HBB), uses Gaussian transform for target localization, and replaces ProbIoU with CIoU loss function to improve the accuracy of detecting rotated targets. To achieve model lightweight and improve detection performance to small targets, we replace the 32-fold downsampling-based feature fusion network with a 16-fold downsampling multi-scale feature fusion network. An improved C2f-PKI module is introduced to enhance multi-scale feature extraction and increase the model’s perception of critical regions and attention to central features. Experimental results show that Ro-YOLOv8-PKI outperforms the YOLOv8n baseline, improving F1 score and mean Average Precision (mAP) by 5.8 % and 9.6 %, respectively, while reducing parameters and model size by 69.1 % and 62.7 %. Additionally, the model achieves mAP gains of 2.3 %, 2.2 %, and 3.1 % over other rotated target detection algorithms, including ROI-Transformer, ReDet, and S2-Anet. This approach offers a practical reference for lightweight rice disease detection in natural environments and presents a new perspective on traditional parallel bounding box-based detection methods. An application has also been developed to demonstrate the real-world applicability of Ro-YOLOv8-PKI in field conditions. Part of the rice blast test dataset used in this study and the sheath blight dataset for future research are available at: https://github.com/qingyun259/RiceLeafBlastDataset.
Why it matches plant phenotyping methodsイネ葉いもち病の症状を画像から検出・定位する軽量アルゴリズムを開発し、精度比較と実環境アプリケーションまで評価しており、植物病害状態の画像ベース表現型取得が中心である。
abstractThis paper proposes a lightweight rotational rice leaf blast detection algorithm named Ro-YOLOv8-PKI.
Breeding salt-tolerant pumpkin cultivars is crucial for improving crop quality and yield. In this study, a high-resolution imaging-based phenotyping platform was developed to capture true leaf images of pumpkin seedlings subjected to salt stress, and plant experts conducted field assessments to determine the severity of salt damage. After image preprocessing, binary mask images were generated, and the maximum likelihood values of normalized intensity were extracted in the red, green, and blue channels to establish a salt stress status index (β) for characterizing stress levels. The β value shows a strong correlation with SPAD value,which indicates that it can effectively reflect the chlorophyll content in leaves, thereby reflecting the physiological changes in leaves affected by salt stress. A leaf texture factor (α) was employed to investigate the directional characteristics of the leaf texture, it can facilitate the effective differentiation of the clusters identified in the clustering analysis and enhance model precision by incorporating detailed leaf structural features. The performance of machine learning, deep learning, and statistical modeling approaches was compared. Statistical model integrating β and α exhibited superior predictive accuracy, with a coefficient of determination, root mean square error, and mean absolute error of 0.901, 0.057, and 0.046, respectively, in the validation dataset. Accuracy assessment among 49 germplasm accessions achieved 95.65 %, demonstrating the model’s reliability. Compared to conventional salt injury assessment, this approach offers higher efficiency and greater objectivity, enabling rapid and accurate identification of salt stress levels in pumpkin seedlings. This study provides a rapid and efficient method for assessing salt stress in pumpkin seedlings, contributing to a deeper understanding of stress response mechanisms and facilitating the selection of salt-tolerant cultivars. Moreover, these findings offer a valuable reference for salt stress identification in other plant species.
Why it matches plant phenotyping methodsカボチャ葉画像から塩ストレス状態を推定する画像ベース表現型取得・解析手法を開発し、専門家評価および49系統で性能検証しており、手法が研究の中心である。
abstracta high-resolution imaging-based phenotyping platform was developed to capture true leaf images of pumpkin seedlings subjected to salt stress
In situ detection of plant ion signals faces technical limitations in terms of real-time capability, minimal invasiveness, and data analysis. Therefore, the development of sensors for in vivo plant detection and construction of time-series prediction models to analyze the dynamic patterns of ion concentrations in plants are imperative. This study presents a microneedle electrode system for potassium ion (K⁺) sensing, which is applied to real-time in situ detection in lettuce. The microneedle ion-selective electrodes (ISEs) fabricated herein exhibited a rapid potentiometric response (within < 15 s), with concentration responses adhering to the Nernst equation. During in vivo plant detection, the system captured instantaneous ion-signal changes upon exogenous application without influencing subsequent plant growth. This study demonstrates the pioneering application of time-series prediction (nonlinear autoregressive neural network model) to analyze in vivo K⁺ signals in lettuce, accurately forecasting ion concentration dynamics over time and identifying the transition pattern from signal fluctuation to stabilization. The integration of microneedle ISE-based in situ plant monitoring with time-series prediction represents a crucial and reliable approach to agricultural sensor innovation, providing a novel paradigm for precision agriculture and plant stress response research.
Why it matches plant phenotyping methods植物体内のK⁺濃度という生理状態を、低侵襲なマイクロニードルISEでリアルタイム取得し、応答性能を検証するとともに時系列予測で解析する手法が研究の中心である。
abstractThis study presents a microneedle electrode system for potassium ion (K⁺) sensing, which is applied to real-time in situ detection in lettuce.
We present the first application of geometry-based relationship constraints for point-cloud registration and unsupervised 3D reconstruction of tree structure in semi-arid forest using unmanned aerial vehicle (UAV) photogrammetry. Accurate three-dimensional (3D) reconstruction of tree structure is essential for a plethora of subsequent tasks like assessing ecosystem health and informing sustainable forest management strategies, in particular over ecologically sensitive arid and semi-arid ecosystems that increasingly face decline due to prevalence of environmental stressors. This highlights the need for high-resolution geospatial monitoring approaches. While UAV-based photogrammetry offers a flexible and cost-effective means of capturing forest structure, conventional top-of-canopy imaging fails to sufficiently represent critical under-canopy features, including stem morphology and lower crown structure. Here, we suggest an integrated 3D reconstruction framework that combines dual-layer UAV photogrammetry, acquiring data from both above and below the canopy, with an innovative geometry-based point cloud registration method. Unlike conventional approaches like Iterative Closest Point (ICP) and Random Sample Consensus (RANSAC), this method leverages spatial relationships among individual trees to robustly align multi-view point clouds acquired under occluded and variable conditions. To further refine the reconstructed tree models, we suggest an updated unsupervised Generative Adversarial Network (Denoise-GAN), enabling both noise reduction and structural completion without reliance on labeled training data. The resulting models were used to extract key phenotypic features with high accuracy compared to reference data (root collar diameter (DRC) R² = 0.93, height R² = 0.97,Crown area R² = 0.99, number of stems R² = 1), providing vital indicators for quantifying forest structure and health. The presented methodology not only enhances the completeness and accuracy of 3D tree reconstruction in semi-arid forest, but also represents a significant advancement toward a scalable, data-driven semi-arid forest monitoring system. This workflow offers substantial potential for ecological applications, particularly in degraded and topographically complex ecosystems.
Why it matches plant phenotyping methodsUAV画像からの3D樹木再構成、点群登録、ノイズ除去・構造補完を開発し、樹木形質の抽出精度を検証しているため、植物フェノタイピング手法が中心である。
abstractWe present the first application of geometry-based relationship constraints for point-cloud registration and unsupervised 3D reconstruction of tree structure in semi-arid forest using unmanned aerial vehicle (UAV) photogrammetry.
In-orchard apple size grading remains challenging under occlusions, variable illumination, and irregular fruit morphology. We present a contact–curvature strategy that integrates bending sensors into fin–ray flexible fingers to estimate fruit size during grasping, unifying grasp–and–grade operations. Neural Radiance Fields (NeRF) reconstruction provides offline ground-truth curvature for calibration, yielding strong linear agreement between sensor readings and true curvature (R2=0.9852). Using temporal curvature features across approach, grasp, and steady phases, a gradient-boosting regression model predicts fruit diameter with R2=0.9577 and RMSE = 1.19 mm on the test set. In laboratory conditions, the system achieved an overall grading accuracy of 98.0 % for 200 apples classified into four grades, with a processing capacity of approximately 6 apples·min⁻¹, meeting real-time requirements. In a small-scale orchard pilot study, the system maintainedR2=0.94 andRMSE=1.27 mm, achieving 96 % grading accuracy versus 77 % for a camera-only approach. Compared with vision-only sizing methods, contact-curvature sensing demonstrates inherent robustness to occlusion and illumination while better tolerating morphological irregularities. A methylene–blue protocol confirmed non–destructive operation. Contact–curvature sensing is robust to occlusions/illumination and can, in principle, extend to other near–spherical crops.
Why it matches plant phenotyping methods果実径という植物器官形質を、接触・曲率センサーと回帰モデルで推定する手法を開発し、校正・精度検証・圃場評価まで行っており、表現型取得法が研究の中心である。
abstractWe present a contact–curvature strategy that integrates bending sensors into fin–ray flexible fingers to estimate fruit size during grasping, unifying grasp–and–grade operations.
Accurate and rapid monitoring canopy-scale nitrogen content (CNC) on pear trees is crucial for precise application of nitrogen fertilizer. Unmanned Aerial Vehicle (UAV)-based spectral analysis is becoming a promising solution for fast monitoring plant nutrition. However, complex data collection conditions, e.g., unpredictable local microclimate, in orchards could easily compromise the quality of spectral images, thereby affecting the estimation accuracy of CNC inversion model. This study aimed to enhance the quality of canopy-scale raw spectral images to improve the accuracy of CNC inversion through the fusion of ground-space spectral imagery. Firstly, collected leaf-scale hyperspectral images, i.e., spectral reflectance and color data, were used as reference values to enhance the quality of canopy-scale raw multispectral images through constructing mapping models based on machine learning algorithms. The conversion of spectral reflectance and color data between leaf-scale and canopy-scale were conducted using the 4SAIL model and the CIELAB color space, respectively. Then, according to the accuracy of mapping models from four classic machine learning algorithms, the RF algorithm was the optimal choice for constructing CNC inversion models. Furthermore, 10 Vegetation Indexes (VIs), 6 Color Indexes (CIs), and their combinations were analyzed using fitting models with simulated canopy-scale spectral reflectance and leaf-scale color data. Based on the top three R² and RMSE values in each type of model, CNC inversion models were constructed using single VI, single CI, and combinations of VIs and CIs. Meanwhile, four methods were tested in each inversion model. The experimental results showed that the R² and RMSE values of the models using mapped data were averagely improved 0.066 and 0.006, respectively, compared to those using raw canopy reflectance and color data. Among all the inversion models using the mapped data, the combination 1 (C1) inversion model (7 VIs and 2 CIs) performed the best, with R², RMSE, nRMSE, and MAE values reaching 0.832, 0.155, 8.333%, and 0.152, respectively. Finally, compared to the C1 inversion model, by screening the inversion results from multi model, the R² of CNC inversion model increased 0.089, enhancing to 0.921. Meanwhile, the RMSE, nRMSE, and MAE decreased 0.038, 2.043%, and 0.072. reaching 0.117, 6.290%, and 0.080, respectively. This study effectively improved the accuracy of CNC inversion by the fusion of ground-space spectral imagery and can offer reference for the application of nitrogen fertilizer in pear orchards.
Why it matches plant phenotyping methodsナシ樹冠の窒素含量という植物形質を、地上・空撮スペクトル画像の融合と機械学習で推定する手法の開発・精度検証が研究の中心である。
abstractThis study aimed to enhance the quality of canopy-scale raw spectral images to improve the accuracy of CNC inversion through the fusion of ground-space spectral imagery.
Accurate identification of plant diseases is critical for maximizing agricultural productivity, particularly in high-value crops like Peruvian coffee, where traditional manual diagnostics remain error-prone and inefficient. While numerous studies have explored convolutional neural networks (CNNs) for disease classification, achieving optimal performance hinges on the precise tuning of hyperparameters a process often relegated to suboptimal trial-and-error methods. Using metaheuristic optimization algorithms to detect such hyperparameters would be a correct approach. For this reason, with the artificial bee colony (ABC) optimization method was goal to determine the hyper-parameters of the proposed CNN architecture in the study. In accordance with this goal, both Peruvian Coffea Dataset (CoLeaf-DB), which is an up-to-date dataset, and Arabica Coffee Leaf Dataset (AcLeaf-DB) which is a reliable dataset with which many studies have been conducted on this subject and which can be benchmarked, were used. On CoLeaf-DB, which is a current dataset used in the study, values of 0.94, 0.94, 0.94, 0.95 were obtained in terms of precision, recall, F1 score, accuracy performance metrics, respectively. Same to order, the values of 0.95, 0.95, 0.95, and 0.97 were obtained from the AcLeaf-DB. When the obtained values are compared with state-of-the-art (SOTA) studies, it is revealed that the determination of hyperparameters with the proposed optimization method and the CNN-based architecture developed on this basis have an extremely important effect on disease detection.
Why it matches plant phenotyping methodsコーヒー葉の病徴を画像から分類するCNNと、ABCによるハイパーパラメータ最適化を開発し、複数データセットで性能比較しているため、病害フェノタイピング手法が中心である。
abstractwith the artificial bee colony (ABC) optimization method was goal to determine the hyper-parameters of the proposed CNN architecture in the study.
Efficient and high-throughput prediction of crop yield and accurate assessment of varietal drought tolerance are essential for modern precision breeding and agricultural resource management and optimization. In this study, we propose a Stages-based Multimodal Data Fusion Model (S-MDFM) by integrating low-cost, high-throughput UAV-based multimodal imagery and multivariate data extracted from images. A staged error metrics (SEMs) propagation mechanism is constructed to capture the dynamic characteristics across growth stages and their dependencies with final yield, thereby improving the accuracy of cross-stage yield prediction (R2 = 0.8536, rRMSE = 16.12 %), validated prediction accuracy using data from the subsequent year in the same cropping area (R2 = 0.8370, rRMSE = 16.63 %). Based on wheat yield under varying water treatment gradients in drought stress experiments, the Drought Stress Tolerance Index (DSTI) and Drought Stress Susceptibility Index (DSSI) were employed to construct a Drought Resistance Index Differential (DRID) evaluation system. This dual-index approach enables multi-level screening of wheat varieties for drought resistance and quantitatively captures the synergistic relationship between yield performance and water adaptability and is capable of performing multi-level screening and quantifying the synergistic relationship between yield performance and water adaptability in wheat varieties, with an identification rate of drought-tolerant and high-yielding cultivars reaching 83.3 %. A multimodal fusion strategy based on discontinuous time-phase observations provides technical support for yield assessment and precise identification of drought-resistant and high-yielding varieties of wheat at different fertility stages, and the method provides a scalable multimodal data integration framework for varietal selection and breeding under drought-stressed field conditions, which is of great practical value for precision agriculture breeding applications.
Why it matches plant phenotyping methodsUAVマルチモーダル画像から画像特徴を抽出し、段階的データ融合モデルでコムギ収量を予測し、乾燥耐性品種を定量スクリーニングする手法が研究の中心であり、翌年データによる検証も行っている。
abstractwe propose a Stages-based Multimodal Data Fusion Model (S-MDFM) by integrating low-cost, high-throughput UAV-based multimodal imagery and multivariate data extracted from images.
Hyperspectral imaging (HSI) has recently emerged as a valuable tool for various agricultural applications. However, the widespread adoption of hyperspectral imaging is hindered due to the high cost and complexity of collecting and processing hyperspectral images. To address this gap, we introduce Agro-HSR,¹1Link to dataset: Agro-HSR. a large-scale RGB to hyperspectral image reconstruction dataset of sweet potatoes, specifically curated to promote easy access to hyperspectral images for the agricultural community. Agro-HSR comprises 1322 pairs of RGB and hyperspectral image cubes from 790 samples across three sweet potato varieties. For 141 of these samples, the agro-product quality attributes are included in the dataset. Each hyperspectral image cube covers 31 evenly spaced bands within the wavelength range of 400–1000 nm. Benchmarks for hyperspectral image reconstruction were conducted to demonstrate the importance and applicability of Agro-HSR. These benchmarks evaluated the ability to predict critical quality parameters in sweet potatoes, including Brix, dry matter, and firmness, from reconstructed hyperspectral images. Agro-HSR enhances the accessibility of hyperspectral images and promotes opportunities for cross-domain research in deep learning and agricultural science, addressing critical challenges in assessing the quality of agro-products.
Why it matches plant phenotyping methodsサツマイモのハイパースペクトル画像再構成データセットを構築し、再構成画像から品質形質を推定するベンチマークを実施しており、植物形質取得・推定手法が中心である。
abstractBenchmarks for hyperspectral image reconstruction were conducted to demonstrate the importance and applicability of Agro-HSR.
Monitoring aboveground biomass (AGB) using high spatial and temporal resolution remote sensing data is important for smart agriculture. Significant technological advances have been made in developing satellites with very high spatial resolution, delivering a promising avenue for vegetation observations. However, the high costs and limited revisit periods of high-resolution satellites hinder their widespread use, leaving the feasibility of combining vegetation indices (VIs) and textures derived from satellite images for AGB estimation uncertain and the quantitative improvements achieved by incorporating textures into estimation unclear. Airborne hyperspectral imaging with high spectral and spatial resolution offers a fresh opportunity to simulate the satellite imaging process objectively and realistically across both spectral and spatial dimensions. The study first evaluated the potential benefits of combining textures and VIs derived from different high-resolution satellites to enhance AGB retrieval. Rice samples and UAV hyperspectral data were collected throughout the rice growth cycle over three consecutive years. Each hyperspectral image was resampled in spectral and spatial dimensions to simulate nine multispectral satellites with sub-meter spatial resolution (WorldView-3, WorldView-2, GeoEye-1, SuperView-1C, GaoFen-2, Beijing-2, Jilin-1, GeoSat-2, KomPast-2). VIs, textures, and their combinations were employed to establish AGB models for the pre-heading, post-heading, and the entire growth stage, respectively. The results showed that combining VIs and textures always achieved the greatest rice AGB estimations, with the integration of multiple satellite data always yielding the best outcomes (overall validation rRMSE ≤ 0.35). For the texture-based monitoring, the impact of satellite spatial resolution was more pronounced on influencing the estimation effectiveness than spectral bands. The monitoring accuracy of rice AGB demonstrated a nonlinear decreasing trend as the spatial resolution dropped, and combining VIs and textures mitigated the negative impact of reduced spatial resolution on the monitoring accuracy of rice AGB. The combination of VIs and textures showed a compensatory effect and combining VIs and textures derived from red-edge band could offset the impact of the reduced spatial resolution on AGB estimation. The involvement of textures in modelling exerted an overall bigger impact on rice AGB estimation than the inclusion of red-edge variables. Satellites with higher spatial resolution and a red-edge band always performed the best in AGB estimation. This study facilitates the optimization of sensor design and farmland management.
Why it matches plant phenotyping methods高解像度リモートセンシング画像からイネの地上部バイオマスを推定する手法を、スペクトル情報・テクスチャ・空間解像度の組合せとして評価・検証しており、植物形質取得が研究の中心である。
abstractThe study first evaluated the potential benefits of combining textures and VIs derived from different high-resolution satellites to enhance AGB retrieval.
Herbicides play a crucial role in cropping systems by providing effective weed control strategies that help farmers eliminate yield-reducing weeds. However, crop injury may result from herbicides applied in current or previous cropping systems, and in some instances, this injury may reduce crop yield. Currently, herbicide related crop injury is commonly determined by subjective visual assessments. Spectral imaging provides an alternative solution, which is high-throughput and non-invasive. In this study, a novel machine vision method utilizing hyperspectral imaging (HSI) and multispectral imaging (MSI) was developed and integrated into Colby’s method—a traditional approach in weed science for analyzing the interaction effects of herbicide mixtures. Mesotrione and diflufenican, both herbicides that cause bleaching symptomology, were applied in this study. Two rounds of field experiments were conducted in the summer of 2024, where hyperspectral and multispectral images were collected 26 DAT in each trial. Partial Least Squares Discriminant Analysis (PLS-DA) models were built to identify soybean injury from mesotrione, diflufenican, and the mixture. For Colby’s method to study the interaction effect, spatial-spectral features were generated from MSI. The HSI models achieved an accuracy exceeding 90 %. Thirteen distinct features were identified and selected to illustrate the synergistic effects of the herbicides, showing consistency across two experimental rounds and aligning with findings from traditional methods.
Why it matches plant phenotyping methods除草剤によるダイズ傷害という植物状態を、HSI/MSIと機械学習で客観的・高スループットに推定する手法を開発し、実験間で検証しているため、植物フェノタイピング手法が中心です。
abstractSpectral imaging provides an alternative solution, which is high-throughput and non-invasive.
Field / plotMultimodalClassificationStress / disease detectionDisease symptoms / severity
In phytopathological diagnostics, traditional unimodal, vision-based methodologies often encounter limitations in performance due to the ambiguous manifestation of disease symptoms and the complexity of background environments. In particular, the robustness and generalizability of such methods are further limited under field conditions characterized by variable illumination, occlusion, and background clutter. Conversely, textual modality exhibits robust semantic encoding capabilities, enabling precise characterization of disease phenotypes and enhancing the discriminative capability of the model. However, textual modality remains underexploited and insufficiently integrated within current research paradigms. In this study, we propose a multimodal diagnostic framework that leverages large multimodal models to automatically generate structured textual descriptions from crop imagery, thereby reducing reliance on manual annotations. To further enhance cross-modal integration, a Projected Visual–Textual Discriminant (PVD) module is introduced. Empirical results indicate that by incorporating generated textual data, most multimodal architectures consistently outperform their unimodal equivalents across diverse visual backbone networks. Notably, the combination of CogAgent with CLIP (ViT-L/14) and the PVD module achieves an F1 score of 70.76%, while LLaVA combined with ResNet50+LSTM achieves 66.38%. The proposed methodology offers a practical and scalable solution for in-field crop disease diagnosis by better aligning visual and textual representations, enhancing classification performance, and obviating the need for manually curated textual annotations by automatically generating descriptions using the Automated Image Description Generation module.
Why it matches plant phenotyping methods作物画像から病徴を推定するマルチモーダル診断フレームワークを開発し、分類性能を比較評価しており、植物病害状態の表現型取得・推定が中心である。
abstractwe propose a multimodal diagnostic framework that leverages large multimodal models to automatically generate structured textual descriptions from crop imagery
This paper presented Bit-STED, a novel and simplified transformer encoder architecture for efficient agave plant detection and accurate counting using unmanned aerial vehicle (UAV) imagery. Addressing the critical need for accessible and cost-efficient solutions in agricultural monitoring, this approach automates a process that is typically time-consuming, labor-intensive, and prone to human error in manual practices. The Bit-STED model features a lightweight transformer design that incorporates innovative techniques for efficient feature extraction, model compression through quantization, and shape-aware object localization using circular bounding boxes for the roughly circular shape of the agave rosettes. To complement the detection model, a novel counting algorithm was developed to manage plants spanning multiple image tiles accurately. The experimental results demonstrated that the Bit-STED model outperformed the baseline models in terms of detection and agave plant count performance. Specifically, the Bit-STED nano model achieved F1 scores of 96.66% on a map with younger plants and 96.43% on a map with larger, highly overlapping plants. These scores surpassed state-of-the-art baselines, such as YOLOv8 Nano (F1 scores of 96.42% and 96.38%, respectively) and DETR (F1 scores of 93.03% and 85.61%, respectively). Furthermore, the Bit-STED nano model was significantly smaller, being less than one-eighth the size of the YOLOv8 nano model (1.4 MB compared to 12.0 MB), had fewer trainable parameters (0.35M compared to 3.01M), and was faster in average inference times (14.62 ms compared to 18.28 ms).
Why it matches plant phenotyping methodsUAV画像からアガベ個体数を推定する軽量検出・カウント手法の開発と性能比較が中心であり、植物個体数という観測可能な形質を抽出している。
abstracta novel and simplified transformer encoder architecture for efficient agave plant detection and accurate counting using unmanned aerial vehicle (UAV) imagery
Better matching of the timing and amount of fertilizer inputs to plant requirements will improve nutrient use efficiency and crop yields and could reduce negative environmental impacts. Deep learning can be a powerful digital tool for on-site, real-time, non-invasive diagnosis of crop nutrient deficiencies. A drone-based RGB image dataset was generated together with ground truthing data in winter wheat (2020) and in winter rye (2021) during tillering and booting in the long-term fertilizer experiment (LTFE) Dikopshof. In this LTFE, the crops were fertilized with the same amounts for decades. The selected treatments included full fertilization including manure (NPKCa+m+s), mineral fertilization (NPKCa), mineral fertilization but no nitrogen (N) application (_PKCa), no phosphorus (P) application (N_KCa), no potassium (K) application (NP_Ca), or no liming (Ca) (NPK_), as well as an unfertilized treatment. The image dataset consisting of more than 3600 UAV-based RGB images was used to train and evaluate in total of eight CNN-based and transformer-based models as baselines within each crop-year and across the two crop-year combinations, aiming to detect the specific fertilizer treatments, including the specific nutrient deficiencies. The field observations showed a strong biomass decline in the case of N omission and no fertilization, though the effects were lower in the case of P, K, and lime omission. The mean detection accuracy within one year was 75% (winter wheat) and 81% (winter rye) across models and treatments. Hereby, the detection accuracy for winter wheat was highest for the NPKCa+m+s (100%) and the unfertilized (96%) treatments as well as the _PKCa treatment (92%), whereas for treatments N_KCa and NPKCa the accuracy was lowest (about 50%). The results were similar for winter rye. In the cross-year and cross-cereal species transfer (training on winter wheat, application on winter rye, and vice versa), the mean accuracy was about 18%. The results highlight the potential of deep learning as a digital tool for decision-making in smart farming but also the difficulties of transferring models across years and crops.
Why it matches plant phenotyping methodsUAV RGB画像から作物の栄養欠乏・施肥状態を推定するデータセットと深層学習モデルを構築・評価しており、植物状態の取得・推定手法が研究の中心である。
abstractDeep learning can be a powerful digital tool for on-site, real-time, non-invasive diagnosis of crop nutrient deficiencies.
The timely and accurate prediction of nitrogen status within crops can provide certain data support for precision fertilization. However, few studies have considered using radiative transfer models to estimate the nitrogen concentration (Cn) of crops. This study is based on the PIOSL-5 model to generate a large number of simulation datasets, namely the PIOSLSD dataset. The successive projections algorithm (SPA) is used to select nitrogen-related features. A crop Cn inversion model based on the PIOSL-5 model is constructed using five models: Extreme Learning Machine, Genetic Algorithm Optimized Extreme Learning Machine, Particle Swarm Optimization Optimized Extreme Learning Machine, Third Generation Non Dominated Genetic Algorithm Optimized Extreme Learning Machine (NSGA-III-ELM), and Bat Algorithm Optimized Extreme Learning Machine. The model is compared with traditional data-driven methods and the accuracy of the model is verified using three datasets: RICE23, LOPEX93, and CALIFORNIA. The results showed that the nitrogen characteristic bands of the PIOSLSD dataset filtered by SPA were 1070, 1150, 1405, 1535, and 1725 nm. The Cn prediction based on the NSGA-III-ELM model, which uses these 5 feature bands as inputs, has the best performance. The determination coefficients of the validation set are 0.814, 0.785, and 0.792, respectively. The Cn inversion model based on the PIOSL-5 model constructed in this article achieves remote sensing prediction of crop nitrogen mechanism model, which has certain mechanistic significance for nitrogen nutrition management of crops and improving nitrogen utilization efficiency.
Why it matches plant phenotyping methods作物葉の窒素濃度という植物形質を、放射伝達モデル・特徴波長・機械学習で推定する手法を開発し、複数データセットで精度検証しており、フェノタイピング手法が中心である。
abstractA crop Cn inversion model based on the PIOSL-5 model is constructed using five models
As an important part of terrestrial ecosystems, the growth and development of plants were regulated by a variety of environmental factors. The rapid development of precision agriculture had put forward higher requirements for real-time monitoring of crop growth environments. Although traditional sensing technology could provide environmental data, it had limitations such as strong invasiveness, large dimensional rigidity, and insufficient long-term monitoring capabilities. The review aimed to systematically review the progress and applications of plant flexible sensors in plant science, highlighting their potential to overcome the limitations of traditional sensors through non-invasive, real-time, and dynamic monitoring of plant physiological and environmental parameters. The review focused on the material systems, fabrication processes, and functional applications of plant flexible sensors. Special attention was given to the roles of conductive polymers, carbon-based materials, and biocompatible substrates in sensor development. Plant flexible sensors, due to their mechanical compliance, functional sensitivity, and energy-efficient operation, offered significant advantages over traditional biosensors. These included in-situ monitoring, long-term operational stability, multi-parameter sensing capabilities, and enhanced adaptability to complex environmental conditions. The reviewed literature demonstrated that plant flexible sensors provided effective and precise monitoring solutions across a wide range of plant physiological processes and environmental conditions. The review provided theoretical guidance and technical reference for the design and application of plant flexible sensors in future agricultural research. The insights gained from this review could facilitate the development of smart agriculture systems, promote advances in plant phenomics, and support sustainable ecological monitoring efforts.
Why it matches plant phenotyping methods植物フレキシブルセンサーによる生理状態の非侵襲・リアルタイム計測を中心に、材料、作製、機能応用を体系的にレビューしており、植物フェノタイピング手法のレビューとして適格。
abstractThe review aimed to systematically review the progress and applications of plant flexible sensors in plant science, highlighting their potential to overcome the limitations of traditional sensors through non-invasive, real-time, and dynamic monitoring of plant physiological and environmental parameters.
Accurate and timely prediction of rice yield is crucial for ensuring food security and optimizing agricultural management. This study proposes a novel QRBILSTM-MHSA model (Quantile Regression-based Bidirectional Long Short-Term Memory Network with Multi-Head Self-Attention) for rice yield prediction, synergizing hyperspectral imaging with multi-modal phenotypic data. The model replaces traditional RNN architectures with BILSTM to acquire bidirectional temporal patterns and permanent dependencies in rice growth cycle. A multi-head self-attention (MHSA) is introduced to weight critical growth factors through parallel subspace analysis, while quantile regression (QR) provides interval predictions, simultaneously estimating average yield and fluctuation ranges. Experimental results demonstrate that the proposed model achieves an R2 of 0.927, a MAPE of 2.21%, and an RMSE of 0.22 tons/ha, significantly outperforming traditional methods such as LSTM, BP-NN, RF, SVR, and ARIMA. At a 95% confidence level, the model achieves a prediction interval coverage probability (PICP) of 98.8% and a percentage of interval width mean percentage (PIWMP) of 0.16, indicating high reliability and robustness. This study highlights the potential of integrating hyperspectral data and deep learning for precise and scalable rice yield prediction, offering valuable insights for agricultural decision-making.
Why it matches plant phenotyping methodsハイパースペクトル画像と表現型データからイネ収量を推定する深層学習モデルを開発し、既存手法との性能比較・検証を行っており、表現型取得・推定手法が中心である。
abstractThis study proposes a novel QRBILSTM-MHSA model (Quantile Regression-based Bidirectional Long Short-Term Memory Network with Multi-Head Self-Attention) for rice yield prediction, synergizing hyperspectral imaging with multi-modal phenotypic data.
Sustainable vineyard management requires precise and efficient application of plant protection products to minimise environmental impact while ensuring plant health. This study presents a variable-rate spraying system that integrates an RGB-D camera with a GPU-equipped edge computing platform to enable accurate, real-time adjustment of spray flow rates in vineyards. A tensor-based representation of RGB-D data is employed to accelerate the entire processing pipeline. Based on this structure, a fast approximate meshing method is applied to rapidly generate 3D meshes from point clouds. To incorporate semantic information from RGB images, an instance segmentation model is used to detect grapevine canopies and trellis posts. The resulting canopy masks are used to isolate the canopy meshes, while the trellis posts serve as reference planes for canopy volume estimation via mesh projection. Based on the computed volume, pulse-width modulation signals are generated to dynamically control spray flow rates. Field experiments were conducted to evaluate the system’s effectiveness and real-time performance. The results demonstrated that the estimated canopy volume is a reliable indicator for regulating application rates. Compared to uniform-rate spraying, the proposed system reduced plant protection product consumption by 57.4% while ensuring adequate droplet coverage. Additionally, the system demonstrated satisfactory real-time performance even on entry-level hardware. Overall, the proposed variable-rate spraying system offers an accurate, real-time, and cost-effective solution for precision viticulture, highlighting its potential for commercial deployment in sustainable vineyard management.
Why it matches plant phenotyping methodsRGB-D画像からブドウ樹冠を分離し、3Dメッシュ投影で樹冠体積という植物形質を推定する技術が中心であり、リアルタイム性能と散布制御への有効性も評価している。
abstractan instance segmentation model is used to detect grapevine canopies and trellis posts
Timely and accurate acquisition of winter wheat yield information is crucial for ensuring food security and formulating agricultural policies. Although deep learning methods have become increasingly prominent in crop yield estimation, they often face challenges in simultaneously capturing both fine-grained local patterns and long-term temporal dependencies in time series data. By utilizing EVI, LAI, and fraction of photosynthetically active radiation (FPAR) from MODIS, along with temperature (TEM) and precipitation (PRE) data from ERA5-Land, we propose a novel dual-branch hybrid model named TCN–Transformer (TCT), which synergistically integrates temporal convolutional network (TCN) and transformer architectures to concurrently capture both localized temporal patterns and long-term dependencies. Bayesian optimization was employed for automated hyperparameter tuning, enabling accurate estimation of winter wheat yield under diverse agricultural management conditions. The experimental results demonstrate that optimal performance is achieved by the proposed TCT model in terms of estimating the county-level winter wheat yields across North China on the test set (R2 = 0.80, RMSE = 645.75 kg/ha). It significantly outperforms the individual temporal models (the TCN, LSTM, and transformer) and other comparative models, including traditional machine learning methods (Ridge, RF, LightGBM, and XGBoost) and an advanced hybrid model (CNN-BiLSTM). Specifically, compared with the individual models, the TCT improved R2 by 0.03 to 0.1 and reduced the RMSE by 29.33 to 156.07 kg/ha. It also outperforms CNN-BiLSTM (R2 = 0.78, RMSE = 668.23 kg/ha), achieving lower errors and more robust bias control. To elucidate the decision-making mechanism of the model, the Shapley additive explanations (SHAP) method was employed to analyze the feature importance values across the study region and the temporal feature weights at 8-day intervals. The results reveal that the EVI is the most representative feature, with the model accurately identifying critical growth stages from T20 (February 26) to T28 (May 1), corresponding to the greening to milk phases, respectively. The feature contribution dynamics were further visualized, revealing a transition from FPAR dominance during early greening (T20–T22) to EVI dominance during jointing (T23–T25), EVI‒PRE interactions during heading-milk (T26–T29), and finally LAI‒PRE dominance at maturity (T30–T32). Furthermore, the one-year leave-one-out cross-validation confirms the robustness of the TCT model, the simulation of yield spatial distribution for unseen years is consistent with the official yield data. Additionally, the proposed interpretability framework not only performed excellently in this study but also demonstrated strong generalizability and flexibility, indicating its broad application potential in other crop types and agricultural domains.
Why it matches plant phenotyping methods冬小麦の収量という植物形質を、マルチソースリモートセンシング時系列から推定する新規TCN–Transformer手法を開発・比較検証しており、形質抽出手法が中心である。
abstractwe propose a novel dual-branch hybrid model named TCN–Transformer (TCT)
Crop models are an integral component in greenhouse control systems, enabling the simulation of plant responses to environmental conditions and facilitating optimal operational decisions for high productivity with low energy use. However, existing crop models often lack transferability beyond their original development conditions. Additionally, cultivar-specific parameterization remains challenging, as some parameters can be empirically determined while others require complex calibration. This study adapted the reduced TOMGRO model to simulate growth and yield for four local tomato cultivars under Shanghai greenhouse conditions. Through Sobol’s global sensitivity analysis and Bayesian optimization, four highly influential parameters were identified and optimized, including growth efficiency (E), maintenance respiration coefficient (rₘ), extinction light coefficient (K), and leaf quantum efficiency (Qₑ). This combined approach provides an effective framework for model calibration, with the calibrated model achieving an average R² > 0.94 for node number, plant dry weight, fruit dry weight, and leaf area index predictions in all cultivars. Model validation using 2023–2024 greenhouse data confirmed model effectiveness for the target variables (average R² > 0.92 for cultivar QX and > 0.88 for LZ), whereas the model showed limitations in simulating mature fruit growth. This calibrated model offers reliable predictions of key growth variables, informing both plant breeding and greenhouse management.
Why it matches plant phenotyping methods作物モデルの感度分析・ベイズ最適化によるパラメータ校正と、植物成長形質予測の検証が研究の中心であり、再利用可能な計算フェノタイピング手法に該当する。
abstractThrough Sobol’s global sensitivity analysis and Bayesian optimization, four highly influential parameters were identified and optimized
Accurate simulation of canopy photosynthesis is essential for predicting dry matter accumulation and crop yield. However, most current crop models overlook the effect of vertical distribution of leaf nitrogen and chlorophyll content on photosynthetic capacity at different canopy layers, resulting in greater uncertainties and weaker mechanistic explanation. Here, we developed a novel canopy photosynthesis model that establishes a bridge between chlorophyll content and photosynthetic nitrogen (PN, defined as total leaf nitrogen minus non-photosynthetic nitrogen) across different canopy heights, and then employs chlorophyll content as a reliable proxy forsimulating photosynthesis. The model was calibrated and validated using data from five field experiments under diverse treatments. Results indicate that leaves at higher canopy positions, receiving more light, contain higher nitrogen content and chlorophyll to support greater photosynthetic rates. The nitrogen extinction coefficient (KN), which characterizes the decline in available of leaf nitrogen, decreases exponentially with increasing LAI, varying among canopy depths, cultivars and growth stages. Chlorophyll shows a stronger correlation with photosynthesis compared to leaf nitrogen. By capturing these dynamics, the model enhances the accuracy of photosynthesis prediction by 60%, particularly correcting the overestimation of canopy photosynthesis and dry matter accumulation during post-flowering. These findings advance the understanding and modelling of canopy-scale photosynthesis in crop models and provide insights for better integration with chlorophyll-related remote sensing data.
Why it matches plant phenotyping methodsキャノピーの光合成をクロロフィル量と垂直窒素分布から推定する新規モデルを開発し、5つの圃場実験で較正・検証しており、植物生理形質の推定手法が研究の中心である。
abstractHere, we developed a novel canopy photosynthesis model that establishes a bridge between chlorophyll content and photosynthetic nitrogen (PN, defined as total leaf nitrogen minus non-photosynthetic nitrogen) across different canopy heights, and then employs chlorophyll content as a reliable proxy forsimulating photosynthesis.
Fruit sizing is a major factor in determining yield while non-destructive monitoring of the fruit skin colour and internal quality attributes on the tree can provide valuable maturity and quality information for precision horticulture. Repeated spectral scanning and fruit sizing data of ‘Braeburn’ apples were collected on the tree from about 60 days after flowering until harvest. Assessed variables were: fruit diameter, dry matter (DMC), soluble solids content (SSC), a normalised difference vegetation index (NDVI) and a normalised anthocyanin index (NAI) analysed using indexed non-linear regression based on an adapted von Bertalanffy model (diameter, DMC, SSC), or a logistic model (NDVI, NAI). The reaction rate constants in the models were estimated in common for all fruit in a selection, while the biological shift factors (Δt) estimated the development stage or fruit maturity per individual fruit. Explained parts (R²ₐdⱼ) range from 85 to 97%. Tree location or crop load treatment only minimally affected the rate constants but did affect the estimated Δt values that describe almost all variation in the data. There is a close relationship between the Δt values for diameter, DMC and SSC but less with those of NDVI and almost none with the NAI. These data support the assumption that there is only one stage of fruit maturity, but it is estimated slightly differently depending on the measured variable. The actual relative growth rate strongly depends on the current size. Understanding apple expansion growth will therefore require a closer focus on the cell production period.
Why it matches plant phenotyping methodsリンゴ果実の非破壊スペクトル測定・果径測定と回帰モデルを組み合わせ、果実の成長・成熟・色素変化を個体ごとに推定する技術的ワークフローが中心であるため、植物フェノタイピング手法の応用として含める。
abstractnon-destructive monitoring of the fruit skin colour and internal quality attributes on the tree can provide valuable maturity and quality information for precision horticulture.
Crop status forecasting by crop model simulations can benefit from assimilating remote sensing observations. When conducting data assimilation (DA) using a common procedure – the Ensemble Kalman Filter (EnKF), arbitrary inflation factors are normally adopted to account for unspecified uncertainties, so as to alleviate filter divergence. Here, we developed a more effective Bayesian methodology, in which the uncertainties were systematically quantified by combining multiple methods in one framework. Its applicability and performance in the EnKF were tested using the crop model GECROS (Genotype-by-Environment interaction on CROp growth Simulator) and the data collected from two years of field experiments for rice. Aboveground biomass (Wₐbₒᵥₑ), grain weight (Wgᵣₐᵢₙₛ), aboveground nitrogen (N) content (Nₐbₒᵥₑ), grain N content (Ngᵣₐᵢₙₛ) and leaf traits like leaf dry weight, leaf N content and leaf area index were measured in the experiments. Using only the observations from the first year, the uncertain parameters in GECROS were calibrated by a Markov Chain Monte Carlo approach, while the parameters in the uncertainty model that describes the errors of crop model simulations were estimated simultaneously. The calibrated model parameters performed well in the validation year, except for the simulated leaf traits (Normalized Root Mean Squared Error (NRMSE) > 0.38). Remotely sensed leaf traits predicted by a Gaussian Process Regression (GPR) model were more accurate (NRMSE < 0.32), with uncertainties of the remote sensing observations estimated from the GPR model itself. Assimilating simulated and predicted leaf traits with their estimated uncertainties into EnKF prevented filter divergence, and the forecast accuracy of crop model improved in the validation year. Compared with simulation without assimilating in-season remote sensing observations, the assimilation procedure led the NRMSE to decrease from 0.37 to 0.20 for whole-season Wₐbₒᵥₑ and Nₐbₒᵥₑ and from 0.39 to 0.20 for the end-season Wgᵣₐᵢₙₛ and Ngᵣₐᵢₙₛ. The updated crop traits of our method also agreed better with the measurements than those of common EnKF with arbitrarily assumed uncertainties and with adjusted inflation factors. The developed method contributes to systematic uncertainty analysis in DA and accurate forecasting of crop growth and yield for smart farming.
Why it matches plant phenotyping methodsリモートセンシングから葉形質を推定し、その不確実性を定量化してデータ同化する計算手法が研究の中心であり、植物形質推定・予測ワークフローとして評価されている。
abstractRemotely sensed leaf traits predicted by a Gaussian Process Regression (GPR) model were more accurate (NRMSE < 0.32), with uncertainties of the remote sensing observations estimated from the GPR model itself.
Traditional methods for estimating wheat seedling area, such as manual grid sampling or ground-based sensors, suffer from low precision, labour intensity, and limited scalability under complex field conditions. To address these challenges, this study introduces a pixel-to-area phenotyping framework that integrates Wheat Seedling Former semantic segmentation with Ground Sample Distance (GSD)-based spatial conversion to achieve high-throughput quantification of wheat seedling coverage and growth vigour. The framework employs a three-step preprocessing pipeline, linear regression-based colour calibration, super-green (ExG) segmentation, and modified anisotropic diffusion filtering, to enhance image quality and suppress noise. The Wheat Seedling Former network incorporates a spatial-channel dual attention module to mitigate background interference and a cross-layer feature pyramid architecture to capture fine-scale morphological traits (e.g., leaf edges, tiller distribution). By aligning RGB and multispectral imagery via geometric correction (holography transformation) and spectral correction (soil-reflection suppression), the framework quantifies six phenotypic indices: seedling coverage area, canopy compactness, NDVI, NDRE, chlorophyll index, and foliage projection coverage. Applied to 160 field plots, the model achieved a Pearson correlation coefficient of 0.942 with ground-truth measurements, demonstrating high accuracy. GSD-based spatial conversion reduced scaling errors to < 3 %, enabling precise area estimation (±0.5 m²) even on uneven terrain. Phenotypic analysis stratified plots into three vigor classes: 35 high-performing (≥90 % canopy closure), 83 medium (60–90 %), and 42 low (<60 %), with high-performing genotypes showing 28 % higher drought tolerance. A software tool (Seedling Phenotype Extractor) automates image annotation, phenotypic calculations, and genotype ranking, reducing phenotyping time by 65 %. This pipeline bridges computational precision and field-scale breeding applications, offering a scalable tool for accelerating the discovery of stress-resilient wheat cultivars through rapid, non-destructive assessment of early-season canopy plasticity.
Why it matches plant phenotyping methodsUAV画像、セグメンテーション、GSD変換、スペクトル補正を統合し、複数のコムギ苗形質を高スループットに推定・検証する方法が研究の中心である。ソフトウェアツールと精度評価も含む。
abstractthis study introduces a pixel-to-area phenotyping framework that integrates Wheat Seedling Former semantic segmentation with Ground Sample Distance (GSD)-based spatial conversion to achieve high-throughput quantification of wheat seedling coverage and growth vigour.
Verticillium wilt (VW) is a highly detrimental disease of cotton that causes significant reductions in yield and fiber quality. Efficient and accurate screening of VW-resistant varieties is essential for cotton breeding and production. However, traditional identification methods, such as manual observation, are inefficient and costly. Unmanned aerial vehicle (UAV) and remote sensing technologies have opened new insights into the screening of field crops for disease-resistant germplasm. This study utilized a UAV multispectral platform to collect data from five growth stages of 150 cotton varieties with different VW resistances. The normalized difference vegetation index (NDVI) was identified as a reliable predictor of chlorophyll levels through hierarchical segmentation analysis. We further compared four deep learning models for chlorophyll monitoring: 1D-CNN, CNN-BiLSTM, CNN-BiLSTM-Adaboost, and CNN-BiLSTM-Attention, with the CNN-BiLSTM-Attention model performing best (R² = 0.92). The optimum model was then used to invert the extent of VW infection using single- and multi-period chlorophyll, and the latter was found to have the best results with the highest R² value of 0.96. Multidimensional clustering of chlorophyll content over multiple periods was used to screen different cotton VW-resistant germplasm, and the ISODATA cluster method outperformed the other three methods (K-means, K means++, and GMM). This study highlights that combining a UAV multispectral platform with an accurate chlorophyll inversion model can enable high-throughput assessment of the cotton VW infection in the field, providing a powerful tool for screening cotton VW-resistant germplasm and thus supporting cotton breeding efforts.
Why it matches plant phenotyping methodsUAVマルチスペクトル計測、クロロフィル推定モデル、感染程度の推定、抵抗性判別を技術的に比較・評価しており、植物表現型取得手法が研究の中心である。
abstractThis study utilized a UAV multispectral platform to collect data from five growth stages of 150 cotton varieties with different VW resistances.
Radiometric infrared (IR) imaging is a valuable technique for remote-sensing applications in precision agriculture, such as irrigation monitoring, crop health assessment, and yield estimation. Low-cost uncooled non-radiometric IR cameras offer new implementations in agricultural monitoring. However, these cameras have inherent drawbacks that limit their usability, such as low spatial resolution, spatially variant nonuniformity, and lack of radiometric calibration. In this article, we present an end-to-end pipeline for temperature estimation and super resolution of frames captured by a low-cost uncooled IR camera. The pipeline consists of two main components: a deep-learning-based temperature-estimation module, and a deep-learning-based super-resolution module. The temperature-estimation module learns to map the raw gray level IR images to radiometric-grade temperature maps while also correcting for nonuniformity. The super-resolution module uses a deep-learning network to enhance the spatial resolution of the IR images by scale factors of ×2 and ×4. We evaluated the performance of the pipeline on both simulated and real-world agricultural datasets composing of roughly 20,000 frames of various crops. For the simulated data, the results were on par with the real-world data with sub-degree accuracy — 0.54∘C mean absolute error (MAE) for ×2 scale factor, and 0.84∘C MAE for ×4 scale factor. For the real data, the proposed pipeline was compared to a high-end radiometric thermal camera, and achieved sub-degree accuracy — 0.81∘C MAE for ×2 scale factor, and 0.81∘C MAE for ×4 scale factor. The results of the real data are on par with the simulated data. We show that our pipeline can compete with high-end thermal cameras in terms of quality and accuracy of the temperature and crop water stress index (CWSI) estimations using affordable hardware, with errors of 1.42% for ×2 and 1.86% for ×4 between the ground truth and the estimated CWSI. The runtime of the pipeline is less than 1sec per frame on a CPU, allowing it to run at video rates. The proposed pipeline can enable various applications in precision agriculture that require high quality thermal information from low-cost IR cameras.
Why it matches plant phenotyping methods低コスト赤外線カメラから植物温度と作物水ストレス指数を推定する深層学習パイプラインを開発し、実データ・シミュレーションおよび高性能熱画像カメラとの比較で精度を検証しており、植物フェノタイピング手法が中心である。
abstractIn this article, we present an end-to-end pipeline for temperature estimation and super resolution of frames captured by a low-cost uncooled IR camera.
Chlorophyll fluorescence parameters (CFPs), especially maximum photosynthetic efficiency of optical system II (Fv/Fm), are the intrinsic photosynthesis probes of crop stress and photosynthetic function. Hyperspectral image (HSI) offers a rapid alternative to traditional pulse amplitude modulation for Fv/Fm, but selecting the uninformative wavelengths reduce accuracy. To address this, a Wavelet Cluster (WCL) method based on Continuous wavelet transform (CWT) was proposed to enhance sensitive wavelengths extraction. Spectral data was preprocessed by Savizky-Golay smoothing (SG) and Multiple scattering correction (MSC), followed by CWT decomposition with bior3.3, gaus4 and meyr wavelet functions to form WCL. Sensitive wavelet coefficients (WCs) were selected using Monte Carlo uninformative variable elimination (MC-UVE), and the Partial Least Squares Regression (PLSR) and Random Forest (RF) modeling methods were established. The results showed that (1) the reflectance spectrum decreased with the increase of Fv/Fm in potato leaf. (2) WCL effectively captured chlorophyll fluorescence spectral features. (3) The WCL-RF model outperformed WCL-PLSR, with a calibration set Rc2 of 0.75, RMSEc of 0.0209, a prediction set Rp2 of 0.73, RMSEp of 0.0225, respectively. This study demonstrates the potential of WCL for accurate Fv/Fm detection, and supports for potato canopy photosynthetic activity assessment.
Why it matches plant phenotyping methodsジャガイモ葉のクロロフィル蛍光指標Fv/Fmという植物生理形質を、ハイパースペクトル画像と新規のウェーブレットクラスタ解析で推定し、モデル性能も評価しているため、フェノタイピング手法が中心である。
abstractTo address this, a Wavelet Cluster (WCL) method based on Continuous wavelet transform (CWT) was proposed to enhance sensitive wavelengths extraction.
The detection of virus infection attacking plants mainly depends on polymerase chain reaction (PCR) testing. Nevertheless, the COVID-19 pandemic, during which the availability of the PCR test was limited, highlights the need for reliable alternative methods for detecting viruses. An effective technique for diagnosing plant disease involves the use of an electronic nose (e-nose) that can detect volatile organic compounds (VOCs) emitted by plants. However, the extensive use of e-noses is limited by the noise that can come from temperature and humidity changes. In order to address this limitation, this research focused on optimising filtering techniques to improve e-nose performance in detecting pepper yellow leaf curl virus (PYLCV) infected chilli plants. The samples were taken from commercial plantations, ensuring that those infected grew in a controlled environment, and ensuring PYLCV detection in diverse conditions. The methods of Fast Fourier Transform (FFT), Discrete Wavelet Transform (DWT), and Savitzky-Golay (SG) filtering were used for the purpose of noise filtering. The optimisation of each filtering technique was performed, such as cutoff frequency for the FFT, the decomposition levels and types of mother wavelets for the DWT, and the polynomial degree and number of windows for the SG filter. The optimisation was performed using a deep neural network (DNN). As a result, the DWT symlet4 level 10 with a specific filter length outperformed the FFT and SG method, with DNN accuracy reaching 97.8% and increasing the accuracy of the unfiltered signal by 5.4%. The result was then validated with other classification models. This proves that with a suitable filtering technique, the e-nose can be a reliable instrument for plant disease detection.
Why it matches plant phenotyping methods植物が放出するVOCを電子鼻で測定し、植物ウイルス感染状態を推定する信号処理・分類手法の最適化と検証が研究の中心であるため、植物病害フェノタイピング手法として収載する。
abstractthis research focused on optimising filtering techniques to improve e-nose performance in detecting pepper yellow leaf curl virus (PYLCV) infected chilli plants.
Agroforestry systems might contribute to balance some of the production, environmental and social challenges associated with agricultural intensification in tropical regions. However, the intricate functional dynamics within agroecosystems, combined with their diverse objectives, make it difficult to maximise productivity and identifying factors that constrain crop yields. Canopy structural traits strongly influence light distribution in agroforestry systems, affecting crop variability in light-use efficiency and productivity. Yet, information on 3D vegetation structure in agroforestry systems remains scarce, despite its potential to provide valuable information to better understand the functional complexity of these systems. Our workflow overcome this limitation by incorporating Terrestrial Laser Scanning (TLS) technology and a voxelization approach for light ray tracing (AMAPVox). In this context, this study aims to determine how TLS data, processed using a voxelization approach, can be applied in multi-strata agroforestry systems to quantify the three-dimensional (3D) distribution of Plant Area Density (PAD) across vegetation strata and the total Plant Area Index (PAI). We used detailed multi-scan voxelized data from 28 experimental plots established in Côte d’Ivoire with different species compositions and planting arrangements to quantify the 3D distribution of PAD and key structural traits (PAI, light attenuation and transmittance) at multiple spatial resolutions. Validation with estimates of light measurements based on hemispherical photographs at tree level showed a high level of concordance (R² > 0.42; p-value < 0.05) in estimating plant area index (PAI). Species composition, rather than planting arrangement, notably influenced the vertical distribution of PAD and canopy structural traits. PAI in the plots ranged from 4.94 m² m⁻² to 22.31 m² m⁻². The proposed approach, using TLS data and a voxel-based methodology, enables high-resolution modelling of canopy structural traits in multi-strata agroforestry systems. These results provide highly detailed measurements of key crop yield indicators, allowing decision support to develop management activities that increase crop production while minimising inputs and maintaining ecosystem services.
Why it matches plant phenotyping methodsTLSとボクセル化によるワークフローを用いて、植物群落の3D構造形質（PAD、PAI、光減衰・透過）を定量化し、光学測定で検証しているため、植物フェノタイピング手法が中心である。
abstractOur workflow overcome this limitation by incorporating Terrestrial Laser Scanning (TLS) technology and a voxelization approach for light ray tracing (AMAPVox).
High precision 3D data is becoming crucial for accurate feature extraction. Acquiring 3D data from plants with different growing patterns and thier growth under different environmental conditions is still a challenging task. The utilization of deep learning techniques can overcome some of these challenges, but these techniques often demand good quality training data for 3D point cloud analysis. One of the main challenges in plant phenotyping is the general lack of annotated 3D datasets available to the research community. Constructing such datasets is particularly difficult due to the complexity of capturing high-quality data that accurately represent the intricate structures and diverse morphologies of plants. The development of robust data sets is critical to advance plant phenotyping, allowing precise quantification of plant traits, and addressing challenges in modern agriculture. However, the lack of high-quality, annotated datasets for complex plant structures, such as wheat, hinders the development of effective methodologies. To address this, we introduce Wheat3D PartNet, a comprehensive repository of 1303 3D point cloud models of wheat (Triticum L.), comprising three cultivars: Paragon, Gladius, and Apogee. The 3D point clouds are reconstructed from RGB images of real plants that were acquired from multiple viewpoints and represent different plant structures at different growth rates. Wheat3D PartNet samples are manually labeled into two parts i.e., ears (wheat spikes) and non-ears (leaves and stems) and that captured in drought and watered conditions. Wheat3D PartNet is designed to support segmentation-based trait quantification tasks such as spike counting, spike length estimation, and stress detection-facilitating more precise yield prediction and enabling early agronomic intervention. Extensive experiments using several state-of-the-art 3D deep learning models validate the dataset's utility and challenge level. The methodology behind Wheat3D PartNet is extensible to other crops, including rice and potato, and is expected to significantly boost the research, understanding, and measurements of plants of interest.
Why it matches plant phenotyping methods植物の3D点群を用いた部位セグメンテーション用データセットの構築・検証が中心で、植物形質の定量化を直接支援するため。
abstractTo address this, we introduce Wheat3D PartNet, a comprehensive repository of 1303 3D point cloud models of wheat (Triticum L.)
Since nitrogen (N) underpins plant vitality by forming proteins, nucleic acids, and chlorophyll, quantifying it through leaf nitrogen concentration (LNC, %) becomes pivotal for growth assessment and precision forestry. However, the three-dimensional structure of the canopy, crown shadow and other background factors complicate the estimation of LNC from crown bidirectional reflectance factor (BRF). To address these challenges, we employ canopy scattering coefficients (CSC) to analyze light behavior within canopies. Accurate estimation of N relies on the association of N with chlorophyll, dry matter, water and canopy structure. To improve the LNC prediction, we developed an enhanced spectral index model called the Difference combined Simple Ratio index (DSR), which improves LNC estimation by minimizing the effects of canopy structure and shadows. Results indicate that the shadow-filtering index methods effectively eliminated shadowed pixels, enhancing the correlation between single-band reflectance and LNC. The CSC-based DSR is the best crown-level LNC estimation for Liriodendron sino-americanum plantation (R² > 0.77, RMSE < 0.7). The estimation model exhibits significant potential for mapping crown-scale LNC distributions in Liriodendron sino-americanum plantations, as well as improving the understanding of the confounding effects of canopy structure and shadows on the LNC estimation.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像とLiDARを用い、樹冠レベルの葉窒素濃度という植物形質を推定するモデルを開発・評価しており、形質取得手法が中心である。
abstractwe developed an enhanced spectral index model called the Difference combined Simple Ratio index (DSR)
To enhance fruit yield and quality, this study focuses on precise pre-harvest ripeness assessment and early disease detection. Addressing the limitations of conventional methods, we propose a multimodal flexible sensing and deep learning-based evaluation framework. The developed flexible optoelectronic in-situ sensing system integrates spectral (410-940 nm, 18 channels) and impedance (100 Hz-10 kHz) detection, allowing conformal attachment to mango surfaces for nondestructive monitoring throughout the growth cycle while collecting spectral, impedance, and physicochemical data. The proposed 1DCNN-ATT-BiLSTM-ATT network employs independent branches to extract local features from each modality, followed by attention mechanisms and temporal modelling for comprehensive feature fusion, achieving 97.5 % accuracy on test sets. Field experiments reveal systematic variations in soluble solid content (SSC), moisture content (MC), and optoelectronic signals during ripening. Correlation and Granger causality analyses underscore the necessity of multimodal fusion. This system supports intelligent harvesting and precision monitoring, advancing agricultural practices toward greater efficiency and sustainability while establishing a technical paradigm for precision agriculture. Future work will focus on improving environmental robustness and cross-cultivar applicability.
Why it matches plant phenotyping methodsマンゴー果実に装着する分光・インピーダンス統合センシングと深層学習による成熟度・品質状態推定を開発しており、植物状態の取得・抽出手法が研究の中心である。
abstractThe developed flexible optoelectronic in-situ sensing system integrates spectral (410-940 nm, 18 channels) and impedance (100 Hz-10 kHz) detection, allowing conformal attachment to mango surfaces for nondestructive monitoring throughout the growth cycle
The early detection of plant diseases is critical for ensuring optimal crop health and maximizing yield. This study presents an AI-driven autonomous robotic system, “AgriScout”, designed for the early identification and mapping of Potato Virus Y (PVY) infections in potato crops. The developed system integrates an electric field robot equipped with RGB cameras and a GPS-RTK module for precise image capture and geolocation of infected plants. The collected high-resolution images are transmitted to a cloud-based server, where a YOLO (You Only Look Once) deep learning model processes them to detect PVY-infected plants. The system generates an infestation map with accurate geospatial coordinates of affected areas, facilitating targeted intervention. Field trials were conducted in Prince Edward Island, Canada, to develop a labeled dataset comprising healthy and PVY-infected plants across different growth stages and environmental conditions. The YOLO model was trained and validated using this dataset, achieving a mean Average Precision (mAP@0.5) of 85%, an F1-score of 0.80, a Precision of 0.85, and a Recall of 0.76 during testing. The model demonstrated robust detection capabilities under varying foliage densities, effectively distinguishing infected plants with high accuracy. The results underscore the potential of “AgriScout” as a scalable, real-time disease detection solution for precision agriculture. By automating disease monitoring and reducing reliance on manual scouting, the system enhances farm productivity, minimizes yield losses, and supports sustainable disease management practices. The integration of robotics and AI in pathogen detection represents a significant advancement in agricultural automation, paving the way for intelligent, data-driven decision-making in modern farming systems.
Why it matches plant phenotyping methods植物のPVY感染状態を画像から検出・地理化するロボット撮像とYOLO解析が研究の中心であり、データセット作成とモデル検証も行っているため、病害表現型の計測手法として適格。
abstractThis study presents an AI-driven autonomous robotic system, “AgriScout”, designed for the early identification and mapping of Potato Virus Y (PVY) infections in potato crops.
Accurate perception in complex agricultural environments is challenging due to significant plant occlusion, primarily from leaves, which hinder data collection and increase uncertainty in robotic operations. Deep-learning-based Next-Best-View (DL-NBV) methods address this by using neural networks to predict information gain (IG) for potential camera views and actively repositioning the camera to maximize data collection with minimal views. However, training DL-NBV models requires extensive IG-labeled data. A self-supervised learning-based NBV method, SSL-Global-NBV, enables robots to collect training data autonomously and improve themselves for global NBV planning. Despite its advantages, SSL-Global-NBV has two key limitations: (1) it requires a fixed number of views, limiting scalability across different plant sizes, and (2) it selects views globally, resulting in inefficient view transitions across the entire view space, reducing trajectory efficiency. To overcome these limitations, this paper introduces SSL-Local-NBV, which incorporates local view planning for scalable and efficient view selection. To prevent redundant visits to the same views, a View Trajectory Network (VTN) was proposed to memorize the view trajectory information of visited views. Comprehensive evaluations in simulation and real-world plant reconstruction demonstrated that SSL-Local-NBV reduced trajectory distance by 56%–70% per reconstruction cycle, achieving 267%–300% higher trajectory efficiency than global NBV methods. Compared to SSL-Global-NBV, SSL-Local-NBV improved plant reconstruction efficiency by 5.2% across varying plant sizes, demonstrating greater scalability. For real plants, SSL-Local-NBV achieved over 80% reconstruction, confirming its feasibility in practical applications. Notably, SSL-Local-NBV fully automated training through self-supervised learning, enabling continuous and lifelong robotic learning.
Why it matches plant phenotyping methods植物の3D再構成を効率化するロボット視点計画手法を開発・評価しており、植物形態の取得が中心的な技術貢献である。
abstractA self-supervised learning-based NBV method, SSL-Global-NBV, enables robots to collect training data autonomously and improve themselves for global NBV planning.
Tomato defect detection and grading based on machine vision are crucial in post-harvest operations, significantly enhancing agricultural product value and market competitiveness. However, accurate segmentation and grading of tomato surface defects remain challenging due to significant intra-class variations, imbalanced defect categories, and especially high manual annotation costs. Therefore, an instance segmentation framework (YOLO-ALDS) was proposed for tomato defect segmentation and grading automatically. The proposed framework included fast dataset preparation based on Active Learning (AL) and segmentation based on improved YOLO-DS. For the dataset preparation part, an uncertainty and diversity-driven active learning (UDAL) strategy was proposed for selecting the most informative defect samples to alleviate the annotation cost and enhance labeling efficiency. For defect segmentation, an improved YOLO11-DS segmentation model is developed by introducing Dynamic Convolution modules in the backbone network, adaptively capturing subtle variations and indistinct boundaries of tomato defects. Moreover, to specifically improve the learning capability for challenging samples with complex and ambiguous morphology selected by the UDAL, a novel SlideLoss function is integrated into the YOLO-DS model, dynamically emphasizing optimization on hard-to-segment instances. Experimental results demonstrate that the proposed YOLO-ALDS reduces manual annotation workload by over 40%, and achieves an mAP@0.5 of 84.1%, surpassing traditional YOLO11 by 0.8%, with notable performance improvements of 1.2%, 1.7%, and 3.7% for white defects, hyperplasia, and cracks, respectively. Compared to mainstream segmentation networks, our approach exhibits significant performance advantages. Furthermore, the developed intelligent tomato grading system based on our model attains practical classification accuracy exceeding 96%, highlighting its promising potential for cost-effective and efficient agricultural automation.
Why it matches plant phenotyping methodsトマト表面欠陥を画像からセグメンテーションし、欠陥状態と等級を推定する手法の開発・評価が中心であり、植物状態の観測手法に該当する。
abstractTherefore, an instance segmentation framework (YOLO-ALDS) was proposed for tomato defect segmentation and grading automatically.
As one of the most widely cultivated and consumed crops, wheat is essential to global food security. However, wheat production is increasingly challenged by pests, diseases, climate change, and water scarcity, threatening yields. Traditional crop monitoring methods are labor-intensive and often ineffective for early issue detection. Hyperspectral imaging (HSI) has emerged as a non-destructive and efficient technology for remote crop health assessment. However, the high dimensionality of HSI data and limited availability of labeled samples present notable challenges. In recent years, deep learning has shown great promise in addressing these challenges due to its ability to extract and analysis complex structures. Despite advancements in applying deep learning methods to HSI data for wheat crop analysis, no comprehensive survey currently exists in this field. This review addresses this gap by summarizing benchmark datasets, tracking advancements in deep learning methods, and analyzing key applications such as variety classification, disease detection, and yield estimation. It also highlights the strengths, limitations, and future opportunities in leveraging deep learning methods for HSI-based wheat crop analysis. We have listed the current state-of-the-art papers and will continue tracking updating them in the following GitHub Repository.
Why it matches plant phenotyping methods小麦のハイパースペクトル画像と深層学習による植物の健康状態・病害・収量推定を対象とする方法論レビューであり、フェノタイピング手法が中心です。
abstractThis review addresses this gap by summarizing benchmark datasets, tracking advancements in deep learning methods, and analyzing key applications such as variety classification, disease detection, and yield estimation.
With the advancement of agricultural modernization, precise plant phenotyping—such as stem-leaf separation—has gained significant importance in the fields of intelligent plant breeding and phenotypic trait extraction. Although deep learning techniques offer superior solutions in the task of complex plant structure segmentation, challenges remain due to insufficient feature representation and low interclass separability. To address this issue, this paper proposes a semantic embedding-guided graph self-attention network for plant stem–leaf separation from 3D point clouds. Specifically, the proposed method is built on an encoder–decoder architecture. The multiscale features are first extracted as the receptive field progressively increases to learn the local geometric representation. Following this, the proposed method constructs a feature enhancement module that integrates graph convolution and self-attention mechanisms. By leveraging graph convolution and the self-attention mechanism, both local and global sets of information are aggregated across multiple scales, capturing intricate geometric and topological relationships to ensure highly descriptive and distinguishing feature representations. Afterwards, we employ a hierarchical decoding structure that combines upsampling and feature fusion to progressively reconstruct high-resolution point cloud feature representations. Finally, the integration of semantic-aware discriminative loss with cross-entropy loss is designed to increase intraclass compactness, interclass separability, and regularization, thereby further strengthening class distinction and segmentation quality. To validate the effectiveness and reliability of the proposed method, experiments were conducted on publicly available Plant-3D and Pheno4D datasets. The results demonstrate that the proposed method achieves superior performance from both quantitative and qualitative perspectives in terms of stem-leaf separation, demonstrating a trend towards outperforming existing methods on the tested datasets, with improvements of 3.97% in precision, 4.35% in recall, 4.3% in the F1 score, 5.23% in the IoU and 7.64% in the mIoU. Additionally, t-SNE visualization and qualitative comparisons further confirm the model’s superiority in feature clustering and structural boundary recognition. Our code is publicly available at https://github.com/Ahaoyang1/3D-SemSeg-RandlAnet.
Why it matches plant phenotyping methods植物の3D点群から茎葉を分離する深層学習手法を開発し、公開フェノタイピングデータセットで性能検証しているため、植物形態の取得・抽出法が中心である。
abstractTo validate the effectiveness and reliability of the proposed method, experiments were conducted on publicly available Plant-3D and Pheno4D datasets.
Citrus Huanglongbing (HLB) is highly contagious, and timely detection and removal of HLB-infected citrus trees is extremely important to prevent its spread. However, the robustness of optical imaging-based models remains limited by the variations in data due to different plant varieties, geospatial conditions, and data collection dates, etc. This study aimed to propose a method for robust HLB detection via transfer learning with multispectral-multicolor imaging. Four lightweight neural networks, namely Yolov7, Yolov7-tiny, Yolov4-tiny, and Mask-RCNN were introduced for citrus HLB disease detection across different datasets. Transfer learning on the Orah mandarin dataset was conducted using the Navel orange dataset for pre-training. The results showed that Mask-RCNN achieved the best performance with an mAP@0.5 of 91.65%. By replacing the backbone of Mask-RCNN with MobileNetV3-large, the model Mask-RCNNV3 was established, with an mAP@0.5 of 93.37% and then used for transfer learning for other datasts. Further optimizing the number of transferred layers and sample size, it revealed the most favorable sample size was 20 per class, and the mAP@0.5 gradually increased at the first 9 layers. Mask-RCNNV3 under the best transfer learning parameters, called Mask-RCNNV3_best, achieved the mAP@0.5 of 93.14% for Orah mandarin, 91.82% for Blood orange and 92.36% for Ponkan, respectively. Compared to the original Mask-RCNN model, the training parameters (Params) and GFLOPs were reduced by 82.95% and 96.57%, respectivley. It demonstrated that a limited amount of labeled data proved sufficient to achieve satisfactory performance across the tested cultivars and growing conditions. The FPS of the model was also improved by 4 times compared to Mask-RCNN, illustrating the potential of the model for edge deployment for practical applications. These findings would bridge the gap between research and practical implementation, reduce costly labeling for model training and provide practical tools for citrus growers to use.
Why it matches plant phenotyping methods柑橘HLB感染状态をマルチスペクトル画像から推定するニューラルネットワークを開発・比較・転移学習で検証しており、植物病害フェノタイピング手法が中心である。
abstractThis study aimed to propose a method for robust HLB detection via transfer learning with multispectral-multicolor imaging.
Fusarium head blight (FHB), which is triggered by fusarium graminearum, drastically reduces wheat yield and quality levels while generating harmful mycotoxins, compromising food security and the health of humans and livestock. Effective real-time detection of wheat FHB in field scenarios remains a critical challenge. Consequently, we present SCS-YOLO, an innovative real-time agricultural disease detection model for wheat FHB detection and severity assessment. Moreover, we successfully deployed it on the low-cost, low-power NVIDIA Jetson Nano embedded platform, achieving low-resource real-time detection. First, we restructured the YOLOv5s backbone network using StarNet, maintaining computational efficiency while obtaining richer and more expressive feature representations. Then, we proposed a novel lightweight CB module to replace the C3 module, further reducing the number of model parameters and the computational scale. Finally, we incorporated a weighted Shape-NWD function, which considers the shapes and sizes of bounding boxes, effectively improving the ability of the model to detect small objects. The results demonstrated that the SCS-YOLO model attained a mean average precision (mAP) of 90.51 % while reducing model parameters and giga floating-point operations (GFLOPs) by 39.97 % and 42.13 %, respectively, outperforming the existing models. Subsequently, the diseased spike rate was calculated, and its coefficient of determination (R²) and root mean square error (RMSE) were 0.90 and 3.48, respectively, effectively quantifying the severity of wheat FHB. Additionally, with an average inference time of only 0.26 s on NVIDIA Jetson Nano, SCS-YOLO exhibited strong potential for rapid detection of wheat FHB on edge devices. In summary, this study offers a dependable, efficient, and accurate solution for wheat FHB detection and assessment. Moreover, SCS-YOLO is designed to be flexible, enabling potential extension to the analysis of other crop diseases or crop types.
Why it matches plant phenotyping methodsコムギ赤かび病の画像検出モデルを開発し、病穂率による病勢を定量評価しており、植物の病害状態を取得・推定する方法が研究の中心である。
abstractwe present SCS-YOLO, an innovative real-time agricultural disease detection model for wheat FHB detection and severity assessment.
Vision Transformers (ViTs) have recently demonstrated promising achievements in various computer vision tasks. However, designing a ViT model architecture requires high-level domain expertise, which can be challenging for new researchers to solve real-life problems, such as those in agricultural imaging. Agricultural datasets often include region-specific patterns influenced by factors such as soil metrics, weather conditions, crop imagery, and spectral signatures, making the design of the manual ViT model a time-consuming and expertise-driven process. To address these limitations, this study proposes a Neural Architecture Search (NAS) leveraging Differential Evolution (DE) algorithm which automates the process of fine-tuning hyperparameters, reducing the reliance on manual intervention and enabling the creation of highly optimized ViT models. The proposed approach is trained and tested on agricultural images dataset of tomato leaves, consisting of ten classes. The experimental results demonstrate the effectiveness of DE-based optimized ViT models by showcasing their ability to handle the unique complexities of agricultural datasets while achieving superior accuracy and reliability in classification tasks.
Why it matches plant phenotyping methodsトマト葉の病害状態を画像から分類するVision Transformerのアーキテクチャ探索手法を開発・評価しており、植物病害表現型の取得・推定が中心である。
abstractthis study proposes a Neural Architecture Search (NAS) leveraging Differential Evolution (DE) algorithm which automates the process of fine-tuning hyperparameters
Potassium is a key element for potato growth and development, playing an important role in enhancing stress resistance, improving tuber quality, and ensuring stable high yields. Therefore, precise and dynamic management of potassium is of vital importance. This study conducted a two-year field experiment on potatoes, setting three irrigation levels (W1: 100 % ETC, W2: 80 % ETC, W3: 60 % ETC) and five potassium fertilizer gradients (K0: 0 kg·ha⁻¹, K1: 100 kg·ha⁻¹, K2: 200 kg·ha⁻¹, K3: 300 kg·ha⁻¹, K4: 400 kg·ha⁻¹). The results showed that the W1K3 treatment achieved the highest potato yield of 59,482.16 kg·ha⁻¹, while excessive potassium application (K4) led to a yield reduction. Subsequently, a critical potassium concentration dilution curve was constructed, and the Measured Potassium Nutrition Index (KNIM) and Theoretical Potassium Nutrition Index (KNIT) were calculated. Under the K3 gradient, the KNI values across all irrigation levels were close to 1, corresponding to the maximum potato yield, with a strong correlation observed between KNIM and KNIT (r > 0.87). Concurrently, multispectral data were collected to construct vegetation indices, extract image texture features, and generate texture indices through random combination of these features. The results indicated that most vegetation indices and some texture features were significantly correlated with KNI (P < 0.05), while all constructed texture indices exhibited extremely significant correlations with KNI (P < 0.01). Different variable combinations were used as input variables to develop KNI prediction models based on machine learning. The optimal performance was achieved by the model integrating spectral information and texture indices with the Random Forest (RF) algorithm, which yielded validation set results for KNIM and KNIT as follows: determination coefficients (R²) of 0.824 and 0.795, root mean square errors (RMSE) of 0.061 and 0.064, and mean relative errors (MRE) of 4.817 % and 5.787 %, respectively. As a theoretically calculated index, the correlation between predicted KNIT values and relative yield further confirmed the validity of KNIT, while the generated potassium nutrition status maps intuitively and clearly demonstrated the feasibility and accuracy of KNIT. This study successfully realized non-destructive and real-time prediction of KNIT through machine learning models, providing a quantitative basis for accurately analyzing the dynamic potassium demand of plants. It holds significant practical implications for promoting the green and sustainable development of the potato industry.
Why it matches plant phenotyping methodsジャガイモのカリウム栄養状態を対象に、マルチスペクトル情報・画像テクスチャ・機械学習を統合してKNIを非破壊かつリアルタイムに推定し、検証しているため、表現型取得・推定手法が中心である。
abstractmultispectral data were collected to construct vegetation indices, extract image texture features, and generate texture indices through random combination of these features.
Current cotton disease detection technologies often rely on complex computations and large parameter counts to achieve high accuracy. However, the trade-off between precision and resource demands constrains their applicability in smart agriculture. To address this limitation, a lightweight Feature Dynamic Reweighting Network (FDRW-Net) is proposed based on the YOLO11 model for cotton disease detection in natural scenes, enhancing detection performance while reducing resource consumption. First, an Enhanced Multi-Scale Spatial Attention (EMSA) mechanism is developed and integrated into a novel Information Sharing Head (ISH). Multi-scale dynamic sensing capability is enhanced and background noise is suppressed through this design. Then, an Adaptive Downsampling Module (ADM), based on ADown, incorporates adaptive weight allocation and input feature weighting, effectively reducing computational costs. Finally, the C3T combining MetaFormer and a Convolutional Gated Linear Unit (CGLU) is introduced into the feature extraction framework. Limitations of convolutional neural networks are addressed and sensitivity to fine-grained features is enhanced by this structure. The performance of the proposed model was evaluated on a self-built dataset containing fusarium wilt, brown spot disease, verticillium wilt, and red leaf blight. Two public datasets were also included in this assessment. Results from the self-built dataset indicate that mAP50 is improved by 3.2 % to 93% compared to the baseline network. Parameters, floating-point operations, and model size are reduced by 46.5 %, 39.7 %, and 43.4 %, respectively, in this evaluation. Experiments on public datasets further validate that the proposed model outperforms existing cotton disease detection methods in overall performance. Based on this improved model, a real-time cotton disease diagnosis system was developed for field monitoring and early warning.
Why it matches plant phenotyping methods綿花の病害状態を自然画像から推定する軽量画像解析モデルを開発し、複数データセットで性能検証しているため、植物病害フェノタイピング手法が中心です。
abstracta lightweight Feature Dynamic Reweighting Network (FDRW-Net) is proposed based on the YOLO11 model for cotton disease detection in natural scenes
The global food crisis, exacerbated by the intensification of crop diseases and pests, poses a significant threat to food security and nutrition. Currently, approximately 350 million people are experiencing extreme hunger, and this number is projected to rise to 943 million by 2025. Consequently, there is an urgent need for effective pest and disease management strategies in agriculture. Traditional identification methods are limited by accuracy, cost, and dependence on human expertise, which hinders timely and efficient pest and disease control. This study investigates the potential of artificial intelligence, particularly deep learning techniques, to enhance the detection and classification of plant diseases and pests. The research focuses on addressing four main challenges: data scarcity, outdated network architectures, computational constraints of terminal devices, and resource and compatibility issues. This paper reviews recent advancements in AI technologies, including few-shot learning, innovative training methods and network architectures, lightweight models, as well as deployment and hardware technologies. Additionally, it discusses the integration of AI in agriculture, highlighting the importance of few-shot learning and the application of new technologies such as Generative Adversarial Networks and Transformers in enhancing pest and disease identification. By providing a comprehensive review of state-of-the-art methods and identifying the unique value of AI in revolutionizing agricultural practices, increasing efficiency, and promoting sustainability, this study makes a significant contribution to the field.
Why it matches plant phenotyping methods植物病害の画像認識・分類に関するAI手法を中心にレビューしており、植物の病害状態を観測・推定する方法論レビューに該当する。
abstractThis paper reviews recent advancements in AI technologies, including few-shot learning, innovative training methods and network architectures, lightweight models, as well as deployment and hardware technologies.
This study investigates lettuce growth under extreme environmental conditions by simulating the weather in six climate zones in a plant growth chamber, including Lleida, Adelaide, Paris, San Luis, Singapore, and Fairbanks. The experiment involved weekly exposure to a new city’s climate, simulating “non-terrestrial weather stress,” which is also motivated from the vantage point of space plant growth and its process-control limitations. These simulated conditions shed light on ‘Climate 2050’, when Earth will probably have harsher and more fluctuating conditions. For the period investigated, the real temperature changes could be reproduced well and in real-time in the growth chamber, the actual rain fall was mimicked, and the lighting period was adjusted to the real sunshine exposure in the respective city. The virtual move of the lettuce plant from between six climates with their own profile in temperature, lighting time, and water is assumed to create stress beyond the variability of a weather change within a single climate. Machine learning models, including linear regression, random forest regression, and boosted decision tree regression, were employed to predict weekly lettuce biomass and yield. This study successfully demonstrated the application of machine learning algorithms for predicting lettuce growth under the given range of six climate conditions. Among the tested models, random forest regression consistently delivered the most accurate and reliable biomass predictions, achieving an R² of nearly 99 % and MAPE of 6 % in all scenarios. By introducing tuned correction factors for conditions like drought stress, fertilisation, and mixed soil composition, the accuracy and flexibility of models are enhanced. This research highlights the value of integrating real-time data with machine learning through a digital twin framework, offering a promising direction for climate-resilient agriculture and space-based plant growth systems.
Why it matches plant phenotyping methodsデジタルツインと機械学習を中核に、レタスのバイオマスおよび収量を予測する再利用可能な計算ワークフローを構築・評価しており、植物形質推定法が中心である。
abstractMachine learning models, including linear regression, random forest regression, and boosted decision tree regression, were employed to predict weekly lettuce biomass and yield.
TomatoGreenhouseLiDAR / point cloudRGB-D / ToFStem / branch2D/3D reconstructionArchitecture / morphology / geometryGrowth / development / phenology
Growth monitoring of tomato plants in large greenhouse environments is critical for quality and efficient production. Stem diameter and elongation are key phenotypic traits for plant growth monitoring. Traditional methods, however, rely on manual operations, which are time-consuming and labor-intensive and do not apply to large-scale greenhouses. Currently, automated image-based methods exemplified by three-dimensional (3D) point cloud technology are among the preferred solutions. Nevertheless, the occlusion of plant structures during the information acquisition process is challenging for practical applications. To address this challenge, this study proposes a novel method for plant stem occlusion inpainting using Deep Reinforcement Learning (DRL). Unlike most existing 3D reconstruction approaches that require depth data from multiple viewpoints, our solution captures 3D point cloud data from a single direction. The DRL model is applied to inpaint the incomplete stem for accurate stem reconstruction and phenotypic measurements. Specifically, our approach consists of two parts, structural completion and stem diameter completion. First, we extract the point cloud of incomplete stems from the RGB-D camera data. Second, we obtain the spatial structure of the stems by inpainting the 3D stem centerline with the DRL model. Finally, we add shape features (stem diameters) by inpainting the two edge lines of the stem occlusion part with the DRL model. For stem inpainted 3D point cloud data, we conducted validation experiments by measuring several commonly used stem phenotypic traits in tomato plants, including stem diameter, stem length, and stem inclination. The experimental results show that the Mean Absolute Percentage Error (MAPE) of the occluded main stem diameter is 9.7%, stem length is 5.7%, and tilt angle is 1%. For the occluded branch stem, the MAPE of stem diameter is 23.1%, stem length is 7.9%, and tilt angle is 1.5%. The accuracy of these measurements for occluded stems is acceptable compared to that obtained from 3D point clouds of unoccluded stems. This highlights the significant potential of using DRL to effectively inpaint occluded 3D point cloud data of plants.
Why it matches plant phenotyping methods植物茎の遮蔽部分を3D点群と深層強化学習で補完し、茎径・茎長・傾斜角という表現型形質の測定精度を検証しており、フェノタイピング手法が中心的です。
abstractTo address this challenge, this study proposes a novel method for plant stem occlusion inpainting using Deep Reinforcement Learning (DRL).
The rising prevalence of Phytophthora diseases in forests highlights the need for rapid, non-invasive detection methods. Early-stage root infections are difficult to detect due to the absence of visible above-ground symptoms, while current diagnostics remain slow and invasive. This study investigated whether hyperspectral leaf reflectance could detect root rot caused by Phytophthora alticola in Eucalyptus benthamii. Nineteen commercially planted families were inoculated, and leaf spectra were collected using an ASD FieldSpec 4 sensor. A machine learning pipeline was developed to identify diagnostic spectral signals. Key wavelengths were identified using permutation importance, a genetic algorithm, and self-attention network (SAN) scores. Spectral signals linked to root rot revealed that infection was correlated with leaf pigment accumulation and moisture stress. Three algorithms, random forest (RF), support vector machine (SVM), and SAN, were trained on hyperspectral data to predict P. alticola infection. The SAN achieved 97 % accuracy on a reduced dataset, which included the diagnostic wavelengths from the feature selection step, surpassing the RF (96 %) and SVM (94 %) models. This study demonstrates hyperspectral sensing as an effective tool for detecting Phytophthora root rot using spectra from the foliage and highlights the application of advanced machine learning techniques for plant disease classification.
Why it matches plant phenotyping methods葉のハイパースペクトル反射から根腐病という植物の病態を推定するセンシングと機械学習パイプラインが研究の中心であり、特徴選択と分類性能も評価しているため。
abstractA machine learning pipeline was developed to identify diagnostic spectral signals.
Wheat is a staple crop that suffers significant yield reductions under drought conditions, especially during the critical reproductive stages. Traditional methods for assessing drought resistance in wheat are often destructive, labor-intensive, and fail to capture the multi-faceted nature of drought tolerance. Vegetation indices serve as effective non-destructive indicators of physiological and biochemical traits. However, the potential of high-throughput spectral indices for quantifying drought resistance traits in wheat have not yet been thoroughly investigated. In this study, we employed an unmanned aerial vehicle (UAV) platform combined with machine learning to assess 206 spectral indices across 52 wheat genotypes at various growth stages under both well-watered and drought conditions. We also evaluated 11 traditional traits to examine their correlations with UAV-based traits. Our study identified 127 spectral indices as drought-related traits and revealed significant correlations between traditional and UAV-based traits. We identified three novel drought-related traits-the Color Index of Vegetation (CIVE), Red-Green-Blue Index (RGBI), and Excess Green Minus Excess Red Index (ExG_ExR)-derived from RGB images and correlated with chlorophyll content, showing strong associations with kernel-related traits. Additionally, we developed an advanced prediction model for yield stability under drought conditions using 17 spectral indices selected through machine learning. A comprehensive evaluation value (D) based on these 17 indices enabled the identification of one highly drought-resistant genotype and 13 drought-resistant genotypes, further validated through field experiments. Our study not only confirms the effectiveness of UAV-based traits in indicating drought tolerance but also provides valuable germplasm for the genetic improvement of drought-resistant wheat.
Why it matches plant phenotyping methodsUAV画像・スペクトル指標と機械学習による乾燥耐性関連形質の定量化および収量安定性予測が研究の中心であり、圃場検証も行っているため。
titleUtilizing UAV-based high-throughput phenotyping and machine learning to evaluate drought resistance in wheat germplasm
The accurate identification of the severity of kiwifruit leaf diseases faces significant challenges due to the high morphological similarity between different disease states and interference from complex environmental factors. To address this issue, we propose a Vision Transformer-based severity grading model for kiwifruit leaf diseases, called KDI-Transformer. This model deeply integrates the global modeling capability of Transformer with the local feature extraction advantages of Convolutional Neural Networks (CNNs). It incorporates three innovative modules: the Multi-Scale Perception Module (MSP), which extracts multi-granularity lesion features using parallel multi-scale convolutional kernels and integrates contextual information at different scales; the Adaptive Feature Transmission Module (AFT), which uses dynamic gating weights to adaptively adjust the inter-layer feature transmission ratio, effectively alleviating the feature attenuation problem in deep networks; and the Local-Global Interaction Module (LGI), which employs an attention mechanism for dynamic calibration of local features under global semantic guidance, significantly enhancing the model’s sensitivity to subtle disease differences. Experimental results demonstrate that KDI-Transformer achieves an accuracy of 89.57 %, significantly outperforming various baseline models, and provides a new solution for precise crop management in smart agriculture.
Why it matches plant phenotyping methodsキウイフルーツ葉の病害重症度という植物状態を画像から推定するVision Transformerモデルを開発・評価しており、病害フェノタイピング手法が中心である。
abstractwe propose a Vision Transformer-based severity grading model for kiwifruit leaf diseases, called KDI-Transformer.
In smart agriculture, the precise acquisition of complete 3D plant phenotypic data is critical for applications such as intelligent breeding and growth monitoring. However, due to equipment constraints, environmental noise, and self-occlusion, the collected 3D point cloud data of plants is often incomplete. This incompleteness significantly hinders key tasks in plant phenotypic analysis, including organ segmentation and surface reconstruction, necessitating effective data completion methods. Supervised point cloud completion methods face challenges due to the inherent incompleteness of collected data and the need for extensive labeled datasets. To address these issues, we propose UnPlantPC, an unsupervised plant point cloud completion model built on a self-supervised encoder–decoder paradigm. To effectively capture regions with complex geometric structures in plant point clouds, the model employs a keypoint down-sampling strategy that integrates Euclidean and cosine distances, ensuring the extracted key points are both representative and directionally informative. Additionally, a geometric-aware attention module enhances feature extraction in these regions, further improving the model’s ability to capture intricate geometric details. To align plant point cloud distributions under self-supervised learning, we introduce a novel Region-Aware Contrastive Distance, which provides accurate supervisory information. This innovation enables the model to deliver more precise completion results. UnPlantPC demonstrates state-of-the-art performance across several metrics on the PlantPCom dataset, achieving a notable 15.73% improvement in CDL2 compared to existing models.
Why it matches plant phenotyping methods植物3D点群の欠損補完という表現型データ取得・復元手法を開発し、データセット上で既存手法と比較評価しているため、植物フェノタイピング手法が中心である。
abstractthe precise acquisition of complete 3D plant phenotypic data is critical
Accurate estimation of comprehensive traits such as yield and quality is crucial for optimizing agricultural management practices across the tomato industry chain. Traditional manual methods are time-consuming, labor-intensive, and prone to errors, reducing estimation accuracy. In contrast, modern intelligent estimation approaches based on multi-temporal spatial and spectral feature fusion offer improved efficiency and accuracy but still face challenges such as non-generalizable segmentation models, asynchronous feature extraction and weak correlations. This study proposes a novel pipeline for estimating yield and quality of greenhouse tomatoes using temporal semantic multispectral (TSM) point clouds. An unsupervised deep learning model was designed to register RGB-D images and multispectral (MS) images collected by an unmanned ground vehicle (UGV) plant phenotyping platform. The digital number (DN) point clouds of tomato organs were reconstructed based on the masks predicted by SegFormer with fusion of multispectral and depth modalities (MSD-SF). These point clouds were then radiometrically calibrated using neural reference field with sparse viewpoints (NeREF-S) to generate accurate reflectance point clouds. Finally, multi-temporal spatial-spectral features of tomatoes were extracted from the TSM point clouds, and random forest regression models were developed to estimate traits such as fruit flavor preference, water content, brix, acidity, brix-to-acid ratio, vitamin C content, single-fruit mass, and single-plant yield. The image registration model achieved high accuracy on the test set, with average structural similarity index measure, peak signal-to-noise ratio and learned perceptual image patch similarity of 0.238, 13.116 dB, and 0.374, respectively. The MS point clouds calibrated by NeREF-S significantly improved the signal-to-noise ratio to 11.56 dB. The average rRMSE for all trait estimations was 9.03 %. The results indicate that the proposed estimation method is efficient and accurate, holding promise to become a new paradigm for estimating the comprehensive traits of greenhouse tomatoes.
Why it matches plant phenotyping methods温室トマトの収量・品質形質を推定するため、UGVフェノタイピングプラットフォーム、マルチスペクトル点群生成、画像登録・放射較正、特徴抽出および回帰推定パイプラインを中心的に開発・評価している。
abstractThis study proposes a novel pipeline for estimating yield and quality of greenhouse tomatoes using temporal semantic multispectral (TSM) point clouds.
Wheat diseases pose a significant threat to global food security by severely reducing crop yields. Rapid, accurate, and reliable identification of wheat infection status, disease types, and disease severity levels is essential for effective disease management. Volatile organic compounds (VOCs), which act as early indicators of plant stress, play a critical role in the early detection and diagnosis of wheat diseases. However, the low concentrations, transient nature, and complex composition of VOCs, combined with the dense canopy structure of wheat plants, present considerable challenges for VOC-based disease identification. To overcome these limitations, this study employed proton-transfer-reaction mass spectrometry (PTR-MS) for the rapid detection of wheat diseases, focusing on mitigating fragment ion interference and mass-to-charge ratio overlap during VOCs spectral characterization. A novel feature recombination method was proposed to improve disease identification accuracy. This approach combines prior knowledge-guided feature recombination with convolutional neural networks for feature extraction, enhancing spectral interpretability and reducing feature redundancy, and enabling rapid VOC-based detection of wheat powdery mildew and stripe rust. Experimental validation demonstrates that the proposed method achieves an accuracy of 90.67% in classifying wheat diseases across different severity levels. Importantly, although this method was designed for wheat disease detection, its framework is adaptable and may be extended to other plant health monitoring applications using PTR-MS.
Why it matches plant phenotyping methodsPTR-MSによる植物揮発性成分の取得と、特徴再構成・CNNによる病害および重症度推定が研究の中心であり、植物の病態を直接評価する手法を開発・検証している。
abstractthis study employed proton-transfer-reaction mass spectrometry (PTR-MS) for the rapid detection of wheat diseases
Hyperspectral imaging systems that operate in the visible-near infrared (VIS-NIR) and short-wave infrared (SWIR) spectral regions are increasingly recognized as practical and effective tools for enhancing crop management. However, hyperspectral systems can have some limitations when focusing on specific spectral ranges, particularly for spatial and spectral resolution. Image fusion techniques combining information from different sensors to enhance hyperspectral data can significantly improve spatial and spectral resolution. Fusion data of image and spectral data from the two HSI cameras (VIS-NIRandSWIR)provide complementary information on plant physiology, biochemistry, and morphology before visible plant stress symptoms. This study presents advancements in hyperspectral image fusion achieved by using two line-scan sensors, one for VIS-NIR (397–1003 nm) and the other for SWIR (894–2504 nm), to detect asymptomatic drought stress in strawberry plants. The images from both hyperspectral imaging systems were aligned based on feature and intensity, combined with various geometric transformations for fusion. The resulting fused hyperspectral cube contained 403 bands covering a broad spectrum from 397 to 2500 nm. Given the vulnerability of strawberry plants to drought, which can significantly affect growth and yield, this study aimed to explore the potential of hyperspectral image fusion for high-throughput detection of drought-stressed strawberry plants. The fused images improved the performance of the PLS-DA detection model, increasing classification accuracy by up to 10 %, achieving 99 % accuracy in the prediction set, and reducing error rates compared to independently generated models.
Why it matches plant phenotyping methodsVIS-NIRとSWIRのハイパースペクトル画像融合を開発・評価し、イチゴの無症候性乾燥ストレスという植物状態を高スループットに検出する手法が研究の中心であるため。
abstractThis study presents advancements in hyperspectral image fusion achieved by using two line-scan sensors, one for VIS-NIR (397–1003 nm) and the other for SWIR (894–2504 nm), to detect asymptomatic drought stress in strawberry plants.
Accurate counting of rice grains plays a critical role in rice breeding and thousand-grain weight measurement. However, the accuracy of existing algorithms is insufficient under conditions of high-density and densely bonded rice grain distribution. To quickly and accurately detect high-density and densely bonded rice grains with as many grains as possible, this study developed a lightweight model based on YOLOv5s. First, we built an efficient lightweight model architecture to obtain small-target location and semantic information of rice grains. Second, we use an omni-dimensional dynamic convolution (ODConv) module to replace some of the convolutions of the backbone network to fully extract feature information. We then introduce the mixed local channel attention (MLCA) mechanism to weigh local features through spatial information, allowing the model to locate and identify dense rice grains accurately. Finally, we use the SIoU loss function to improve the convergence speed and accuracy of model training. The model’s detection accuracy was verified via ablation experiments. The results indicated that compared with the original YOLOv5s network, the model size, parameters and floating-point operations per second (FLOPS) of the improved model decreased by 64.16 %, 70.8 % and 28.3 %, respectively, while mAP₀.₅:₀.₉₅ increased by 7.21 %. The mean error rate and mean detection time of the improved model were 0.234 % and 25.9 ms, respectively. Its superior capacity against other detection algorithm models at rapidly detecting densely bonded rice grains. Furthermore, an android application was further developed. After comparative testing on three types of mobile phones, the application was able to effectively and accurately detect and count rice grains, providing an effective solution for rice grain detection and counting.
Why it matches plant phenotyping methods米粒の検出・計数という植物形質取得を目的に、YOLOv5s改良モデルを開発し、アブレーション実験と比較評価、モバイルアプリ実装まで行っており、フェノタイピング手法が研究の中心である。
abstractAccurate counting of rice grains plays a critical role in rice breeding and thousand-grain weight measurement.
Leaf disease recognition is critical for guaranteeing rice quality and yield. However, hindered by similar symptoms and complex background interference in practical field scenarios, existing models face the significant challenge of balancing accuracy and lightweight requirements for edge devices. To address the challenge, this paper proposes a filter sensitivity-based lightweight network (FSLNet), which comprises three key modules: a filter sensitivity evaluation algorithm (FSEval), a sensitive channel spatial attention mechanism (SCSAM), and a sensitivity-driven model compression method (SDMC). To overcome similar symptoms, FSEval is designed to calculate each filter’s sensitivity to different diseases. Then, the channel penalty is imposed on non-sensitive filters (CP-NSF) to make FSLNet extract differential features. To mitigate background interference, during each training epoch, SCSAM can adaptively infer an attention map only from sensitive channels containing differential features, to optimize their weight distribution. To accommodate edge devices in the practical field, SDMC incorporates a sensitivity-driven channel penalty pruning (SDCPP) strategy and a fine-tuning method with a low pruning teacher model via knowledge distillation (FT-LPT-KD) to build a lightweight model. Experimental results on the self-built dataset and public Paddy Doctor dataset demonstrate that FSLNet has only 0.38M parameters, just 1/6th of the smallest benchmark GLDCNet, while can achieve an average accuracy of 95.26% and 97.98%, respectively, which is 1.84% and 0.77% higher than those of the state-of-the-art schemes. Further testing on a resource-limited edge device reveals that FSLNet not only outperforms existing methods in recognition accuracy but also can achieve a real-time recognition speed of 28 FPS.
Why it matches plant phenotyping methodsイネ葉の病徴を画像から認識する軽量モデルを開発し、公開・自作データセットおよびエッジデバイスで性能検証しているため、植物病害状態のフェノタイピング手法が中心である。
abstractthis paper proposes a filter sensitivity-based lightweight network (FSLNet)
Accurate estimation of total leaf area (TLA) is essential for assessing plant growth, photosynthetic activity, and transpiration, but remains a challenge for bushy plants like dwarf tomatoes. Traditional destructive methods and imaging-based techniques often fall short due to labor intensity, plant damage, or the inability to capture complex canopies. This study evaluated a non-destructive method combining sequential 3D reconstructions from RGB images and machine learning to estimate TLA for three dwarf tomato cultivars—Mohamed, Hahms Gelbe Topftomate, and Red Robin—grown under controlled greenhouse conditions. Two experiments, conducted in spring–summer and autumn–winter, included 73 plants, yielding 418 TLA measurements using an “onion” approach, where layers of leaves were sequentially removed and scanned. High-resolution videos were recorded from multiple angles for each plant, and 500 frames were extracted per plant for 3D reconstruction. Point clouds were created and processed, four reconstruction algorithms (Alpha Shape, Marching Cubes, Poisson’s, and Ball Pivoting) were tested, and meshes were evaluated using seven regression models: Multivariable Linear Regression (MLR), Lasso Regression (Lasso), Ridge Regression (Ridge-Reg), Elastic Net Regression (ENR), Random Forest (RF), extreme gradient boosting (XGBoost), and Multilayer Perceptron (MLP). The Alpha Shape reconstruction (α = 3) combined with XGBoost yielded the best performance, achieving an R² of 0.80 and MAE of 489 cm², with significant results across other model combinations. Results were lower when using data from different experiments as train and test datasets (R² = 0.56 and MAE = 579 cm²). Feature importance analysis identified height, width, and surface area as the most predictive features. These findings demonstrate the robustness of our approach across variable environmental conditions and canopy structures. This scalable, automated TLA estimation method is particularly suited for urban farming and precision agriculture, offering practical implications for automated pruning, improved resource efficiency, and sustainable food production.
Why it matches plant phenotyping methodsRGB画像からの3D再構成と機械学習により植物の総葉面積を推定する手法を開発・評価しており、表現型取得が研究の中心です。
abstractThis study evaluated a non-destructive method combining sequential 3D reconstructions from RGB images and machine learning to estimate TLA
Grapevine Leafroll Disease (GLD) poses a significant economic burden on the wine industry in major wine-producing regions. Conventional methods of phenotyping GLD are inefficient and delay vineyard management decisions. The emergence of affordable Unmanned Aerial Vehicles (UAV) provides unprecedented opportunities for GLD high-throughput phenotyping. However, detecting GLD-infected grapevines at the canopy level using UAV images is still a challenge due to the subtle differences between GLD canopy features and background. In this paper, we propose a GLD Detector (GLDD) for mapping GLD epidemics from UAV images. A new attention mechanism module, namely, Channel Attention with Transformers (CAT) is proposed to alleviate the difficulty of extracting high-resolution features from the elongated canopy. We redesigned YOLOv7-tiny for GLDD and conducted a series of ablation experiments to evaluate its performance. Experimental results show that GLDD outperforms YOLOv7-tiny by 3.1% and YOLOv6-tiny by 6.5%, reaching an accuracy of 88.2%. In comparison to several one-stage object detectors such as YOLOv5, YOLOX, PP-YOLOE, and YOLO-FaceV2, GLDD obtains the best detection results. Additionally, compared to convolutional-based detectors such as Faster-RCNN and transformer-based detector SwinT, GLDD performs better in speed and accuracy. Furthermore, GLD-infected grapevine distribution is also mapped by using GLDD detection results at the field scale.
Why it matches plant phenotyping methodsUAV画像からブドウ樹の葉巻病感染状態を推定する検出手法を開発・比較評価しており、植物病害状態のフェノタイピングが中心です。
abstractConventional methods of phenotyping GLD are inefficient and delay vineyard management decisions.
Grapevine winter pruning is a labor-intensive and repetitive process that significantly influences grape yield and quality at harvest and produced wine. Due to its complexity and repetitive nature, the task demands skilled labor that needs to be trained, as in many other agricultural sectors. This paper encompasses an approach that targets using a robotic system to perform autonomous grapevine winter pruning using a vision system and artificial intelligence. In our previous work, we presented a 2D neural network that segmented images of grapevines into 5 different classes of plant organs during their dormant season. In this paper, we expand into the third dimension, introducing point clouds into our algorithm. The 3D approach creates instance-segmented point clouds using depth images and segmentation masks obtained with our 2D neural network. After the 3D reconstruction, the system extracts thickness measurement and uses agronomic knowledge to place pruning points for balanced pruning. The study not only delineates the integration of 2D and 3D methods but also scrutinizes their efficacy in pruning point identification. The real-world performance of the created system was evaluated and statistically analyzed on data collected during field trials in the winter pruning season 2022/2023, where the system was used in a potted vineyard to prune a set of test vines, where the positive success rate is 54.2%. Moreover, as one of the main contributions, the paper underscores a unique facet of adaptability, presenting a customizable framework that empowers end-users to fine-tune parameters according to the expected balanced pruning. This adaptability extends to variables such as the number of nodes to retain on pruned spurs and the preferred cane thickness, encapsulating the versatility of the 3D approach.
Why it matches plant phenotyping methods2D画像分割と3D点群再構成を統合し、ブドウ樹器官の厚さを抽出して剪定点を生成・評価する手法が研究の中心であり、植物形質の取得と技術性能検証を含む。
abstractThe 3D approach creates instance-segmented point clouds using depth images and segmentation masks obtained with our 2D neural network.
Rapid, accurate, and non-destructive estimation of crop water use efficiency (WUE) at the field scale is crucial not only for evaluating water efficient cultivars and practices in scientific research but also for optimizing irrigation schedule in agricultural production. The current lack of efficient methods for high-throughput phenotyping WUE hinders development of sustainable agriculture under globally intensified water scarcity. This study aimed to utilize unmanned aerial vehicle (UAV) multisensory remote sensing data combined with a process model to achieve rapid WUE determination via accurate daily-scale evapotranspiration and aboveground biomass (AGB) estimates. First, vegetation indices, canopy temperature, and canopy structural parameters were extracted from multispectral (MS), thermal imaging (TIR), and radar data and combined with an automated machine learning (AutoML) for AGB estimation. The beta function was then employed to accurately estimate AGB accumulation at a daily step (AGBdₐᵢₗy) over the entire growth period. The daily evapotranspiration (ETdₐᵢₗy) was calculated by the surface energy balance algorithm for land (SEBAL) model driven by MS, TIR, and meteorological data. Finally, the WUE was determined by the ratio of AGBdₐᵢₗy to ETdₐᵢₗy. Multisensory data fusion and further integration with process-based model proved effective for simultaneously estimating AGBdₐᵢₗy, ETdₐᵢₗy, and WUE with R² values of 0.71, 0.93, and 0.79, respectively. Notably, the proposed WUE estimation method can capture different temporal pattern between cultivars with different levels of tolerance to drought. We applied this approach to screen water efficient cultivars and found that appropriate reduction of irrigation can improve WUE. In conclusion, this study shows promising perspective in the use of a UAV-based approach integrating multisensory data with SEBAL evapotranspiration modeling for monitoring and evaluating water consumption and utilization in maize.
Why it matches plant phenotyping methodsUAVマルチセンサーデータとモデルを統合し、トウモロコシのAGB、蒸発散、WUEという植物形質・状態を推定する方法を開発・評価しており、表現型取得が研究の中心である。
abstractThe current lack of efficient methods for high-throughput phenotyping WUE hinders development of sustainable agriculture under globally intensified water scarcity.
The use of multiple camera technologies in a combined multimodal monitoring system for plant phenotyping offers promising benefits. Compared to configurations that rely on a single camera technology, cross-modal patterns can be recorded that allow a more comprehensive assessment of plant phenotypes. However, the effective utilization of cross-modal patterns depends on image registration to achieve pixel-precise alignment - a challenge often complicated by parallax and occlusion effects inherent in plant canopy imaging. In this study, we propose a novel multimodal 3D image registration method that addresses these challenges by integrating depth information from a time-of-flight camera into the registration process. By leveraging depth data, our method mitigates parallax effects, facilitating more accurate pixel alignment across camera modalities. Additionally, we introduce an automated mechanism to identify and differentiate various types of occlusions, thereby minimizing registration errors. To evaluate the efficacy of our approach, we conduct experiments on a diverse dataset comprising six distinct plant species with varying leaf geometries. Our results demonstrate the robustness of the proposed registration algorithm, showcasing its ability to achieve accurate alignment across different plant types and camera compositions. Compared to previous methods our approach is not reliant on detecting plant-specific image features, making it suitable for a wide range of applications in plant sciences. Moreover, the registration approach can scale to arbitrary numbers of cameras with varying resolutions and wavelengths. Overall, our study contributes to advancing the field of plant phenotyping by offering a robust and reliable solution for multimodal image registration.
Why it matches plant phenotyping methods植物フェノタイピング向けのマルチモーダル3D画像位置合わせ手法を開発・評価しており、表現型取得ワークフローの技術的中心である。
abstractwe propose a novel multimodal 3D image registration method that addresses these challenges by integrating depth information from a time-of-flight camera into the registration process.
Fusarium head blight (FHB) is one of the most serious wheat diseases and mainly infects the ear, affecting the yield and quality of wheat worldwide. Segmentation of FHB infection in wheat ear based on unmanned aerial vehicle (UAV) images is feasible and significant in ensuring timely control measures and maintaining food security. The high flight altitude of UAV allows for rapid image acquisition but results in blurred textures and details, and the variability of field environment leads to missed and false segmentation. To address these problems, we first executed the super-resolution (SR) of high-altitude UAV images, and then FHB infection was segmented using a deep gate network. Specifically, an SR network called hierarchical context aggregation network (HCAN) was developed to generate clear textures and detailed characteristics of wheat efficiently through the successive fusion of various contexts. HCAN was superior to the current state-of-the-art methods with a peak signal-to-noise ratio of 29.056 dB and a structural similarity index of 0.9142. Meanwhile, a reception enrichment gate network (REGN) was applied to segment FHB infection in wheat ear through the integration of dual-gate mechanism and multi-scale convolution. REGN gained superior results to those of other segmentation networks with a mean intersection over union of 77.93 %, mean pixel accuracy of 87.43 %, and mean Dice coefficient of 87.06 %. Indistinct edges, missed segmentation, and false segmentation were dramatically alleviated in high-density, overlapping, shaded and overexposed wheat because local and neighboring gate operations enhanced the representation and reception field, and multi-scale convolution could enrich the reception diversity. In sum, the proposed approach provided a reliable, efficient, and accurate determination of FHB infection in wheat on the basis of UAV images and could be extended to the analysis of other diseases or crops.
Why it matches plant phenotyping methodsUAV画像からコムギ穂のFHB感染状態を抽出する超解像・セグメンテーション手法を開発し、性能評価しており、植物病害表現型の取得方法が中心である。
abstractREGN gained superior results to those of other segmentation networks with a mean intersection over union of 77.93 %, mean pixel accuracy of 87.43 %, and mean Dice coefficient of 87.06 %.
Phenotyping, the measurement of attributes or traits, is crucial in selecting superior cultivars for specific environmental situations. This is a time-consuming process when applied to large populations but can be accelerated through the use of deep learning, resulting in an algorithm that can phenotype images of specimens in negligible amounts of time. The primary issue with deep learning is the large quantities of high-quality training data required to make a viable phenotyping pipeline. To address this, we present a semi-synthetic training data generation system which significantly reduces the amount of human effort spent on data collection. We use active learning alongside this system to create DeepCanola, an instance segmentation model that successfully segments and measures the valves from Brassica napus pods. We demonstrate that the model accurately estimates the effect of different winter cold treatments on a range of different cultivars and crop types as effectively as manually curated measurements. Furthermore, the resulting model is effective on data from various experimental settings and on different, but related, species such as Arabidopsis thaliana, Allaria petiolate (garlic mustard) and Raphanus raphanistrum subsp. sativus (radish). This robust tool could be easily scaled, thereby accelerating breeding or fundamental research programs. Code and model weights: https://github.com/kieranatkins/deepcanola.
Why it matches plant phenotyping methods植物の莢画像からバルブを分割・測定する深層学習フェノタイピング手法を開発し、半合成データとアクティブラーニング、複数条件・種での性能検証を行っているため。
abstractWe use active learning alongside this system to create DeepCanola, an instance segmentation model that successfully segments and measures the valves from Brassica napus pods.
For most digital agriculture applications, such as in-season yield predictions, information on crop phenology is a prerequisite. Phenology is largely determined by environmental factors, e.g., temperature, precipitation, and global radiation. Consequently, weather data can be used to predict phenology. Here, we introduce the R package Dynamic Multi-Environmental Phenology (DyMEP) that facilitates such predictions. DyMEP was trained for ten crops, among others winter wheat, spring wheat, barley, green peas, beans and oat, with a large dataset representing the Central European climate. DyMEP fills the gap between complex, highly parameterized crop growth models that are difficult to use by non-experts, and extremely simplified models such as the Growing Degree Day approach. By carefully selecting the environmental covariates to use for each phenological phase, the user can reach suitable prediction accuracy for most applications in DyMEP. If temperature, precipitation, relative humidity, and global radiation are available as covariates to select from, the package achieves absolute errors ranging from 0 to 6 days across all applied phenology phases and root mean square errors ranging from 7 to 17 days on an independent test set. Combining DyMEP-based phenology predictions with ground-based or remote sensing observations holds promise to facilitate digital agriculture applications such as large-scale yield forecasting or monitoring of fields for crop insurance.
Why it matches plant phenotyping methods作物の生育ステージ（フェノロジー）を気象データから予測するRパッケージを開発・評価しており、植物状態の計測・推定手法が研究の中心である。
abstractHere, we introduce the R package Dynamic Multi-Environmental Phenology (DyMEP) that facilitates such predictions.
Nitrogen (N) management is one of the main factors enhancing potato productivity and promoting sustainable agricultural practices. The Nitrogen Nutrition Index (NNI, obtained as the ratio of actual plant N, to the critical plant N concentration) is widely applied to assess the N status of various crops. Traditionally, NNI is calculated using field data, but remote sensing (RS) technologies can offer more rapidly and timely assessment of the spatiotemporal (within field) variability of this index. This study employs multispectral data acquired via Unmanned Aerial Vehicle (UAV) and machine learning (ML) models to estimate potato NNI. A Bayesian hierarchical partially pooled method was fitted to a three-year field experiment in Denmark and extensive ground-based potato datasets to model the critical nitrogen dilution curve (CNDC) and calculate the NNI. Multispectral UAV data were processed to extract four spectral bands and calculate several vegetation indices, which were used as predictors to train and test six ML models: Linear regression, support vector machines, gaussian process regression, stepwise linear regression, ensemble trees and neural networks. Among the compared models, gaussian process regression outperformed, showing R² equal to 0.83 and a RMSE of 0.10 and providing accurate NNI predictions, comparable to ground-based Bayesian estimates. The variability of the NNI was analyzed over the seasons using 28 NNI maps derived from UAV surveys at spatial resolution of 0.04–0.09 m/pixel, capturing spatial variations in crop N status over time. The proposed framework, designed for NNI prediction at the intra-field scale, has the potential to be adapted to different environments and crops. The framework can support practical decisions for precision N management, reducing the environmental impact of potato cultivations and enhancing sustainability.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像と機械学習により、ジャガイモの窒素栄養状態（NNI）を推定する手法を開発・比較・検証しており、植物表現型の取得が研究の中心である。
abstractThis study employs multispectral data acquired via Unmanned Aerial Vehicle (UAV) and machine learning (ML) models to estimate potato NNI.
Tea leaf blight (TLB) is a common tea disease, and accurate detection of the different stages of TLB helps in tea disease control. The color and shape of TLB spots at different stages vary greatly and are easily confused with complex backgrounds; hence, the accuracy of existing methods for detecting TLB at different stages is not high. In this study, a dual-branch TLB detection network (DBTDNet) combining spatial domain and frequency domain information was designed for the accurate detection of TLB at different stages. The dense depthwise separable (DDS) module and wavelet-based feature extraction (WBFE) Bottleneck were introduced into the spatial feature extraction (SFE) branch and frequency feature extraction (FFE) branch of DBTDNet to extract spatial domain features and frequency domain features, respectively, and enhance the localization and recognition of TLB spots at different stages. A multiscale wavelet transform convolution (MSWTC) module was also added to the FFE branch to separate the multi-scale frequency information and obtain clearer shape and texture features of TLB spots. A linear layer was introduced between the dual-branch structures to reduce the gradient information loss. In addition, to better capture TLB spots of different sizes, this study designed a multidimensional neural network (MNNet) structure in the feature fusion part of DBTDNet for fusing the information of different scale feature maps from the output of the dual branch. The experimental results showed that the proposed DBTDNet could more accurately detect TLB spots at different stages than the existing state-of-the-art network models. The mAP@0.5 values of its detection results for yellow TLB spots in the early stage, white TLB spots in the middle and late stages, and total TLB spots were 74.5%, 75.3%, and 75%, respectively, which were 13.2%, 7.5%, and 10.5% higher, respectively, than the baseline model YOLOv9 detection results.
Why it matches plant phenotyping methods茶葉の病斑という植物の病害状態を画像から検出・認識する深層学習手法を開発し、既存モデルと性能比較しているため、植物フェノタイピング手法が中心である。
abstracta dual-branch TLB detection network (DBTDNet) combining spatial domain and frequency domain information was designed for the accurate detection of TLB at different stages.
Wheat lodging is a usual agricultural disaster in wheat growth. It reduces the grain yield and harvesting efficiency. Existing segmentation methods cannot achieve satisfactory performance and trade-offs between accuracy, inference time, and lightweight when facing the challenge of multiple lodging scenes. Therefore, developing an innovative segmentation algorithm that is real-time and low-complexity to identify lodging situations is of great value for improving agricultural production. To achieve these goals, we propose a lightweight and efficient lodging semantic segmentation model, WLUSNet, to separate the lodging area of unmanned aerial vehicle (UAV) images. Inspired by the mixed depth-wise grouping convolution (MC) and the channel feature pyramid (CFP) modules, a multiscale backbone (MC-CFP) is designed to reduce information loss in feature extraction. Then, drawing on the characteristics of MC and the channel attention (CA) mechanism, a space pyramid module (MC-SP) is designed to enhance feature representation by obtaining the global information on the channel feature and the local information on the space feature. To reconstruct a high-resolution feature map, a feature fusion module (EDFF) between the shallow and deep features is introduced to improve segmentation accuracy. The comprehensive experimental results demonstrate that WLUSNet performs excellently well compared with 11 other state-of-the-art (SOTA) segmentation algorithms. WLUSNet achieves a mean intersection over union (mIoU) of 86.9, a mean pixel accuracy (mPA) of 93.26, a model size of 4.1 M, and an inference speed of 26.94 FPS on the self-built UAV remote sensing dataset in this paper. The generation experiment indicates that WLUSNet has the potential to segment other lodging crops, and can provide technical support for segmentation tasks in crop lodging.
Why it matches plant phenotyping methodsUAV画像からコムギの倒伏状態を抽出するセグメンテーション手法WLUSNetの開発と、他手法との性能比較・検証が研究の中心である。
abstractwe propose a lightweight and efficient lodging semantic segmentation model, WLUSNet, to separate the lodging area of unmanned aerial vehicle (UAV) images.
Advanced phenotyping techniques are required in the breeding and management of maize, which is crucial for global food security. Traditional in situ three-dimensional (3D) field phenotyping entails labour-intensive data acquisition. Light detection and ranging technology offers high-resolution maize canopy point clouds under outdoor field conditions, establishing a technical foundation for automated phenotypic trait extraction. However, accurately segmenting individual plants from dense and structurally complex canopy point clouds for single-plant trait analysis is challenging. To address this challenge, we propose a novel framework named Paired-Attention Central Axis Aggregation Network (PACANet) for 3D point cloud-based plant segmentation. Firstly, a 3D paired-attention backbone network is introduced to enhance point-wise feature representations by integrating spatial and channel information, thereby enabling effective learning of high-dimensional point cloud features. Secondly, a projection-based central axis aggregation strategy is incorporated to guide instance separation by projecting plant point clouds onto their respective central axis skeletons, which improves the spatial coherence of segmentation. Additionally, a simulation-based point cloud generation approach is proposed to reduce reliance on large-scale manual annotations, facilitating model training in scenarios with limited real-world population data. Comprehensive experimental evaluations across multiple datasets demonstrate that PACANet consistently outperforms existing plant population segmentation methods. Notably, when trained solely on simulated data, PACANet achieves a state-of-the-art average precision of 0.9246. Finally, based on the segmentation results, phenotypic traits at both the individual plant and organ levels are analyzed under various planting densities, including the field-level distributions of plant height, plant width, leaf base angle and leaf inclination angle, all of which exhibit strong consistency with the validation data. These results highlight the potential of PACANet as a robust and scalable solution for high-throughput phenotyping in smart breeding and precision agriculture. This study provides a new tool for smart breeding and precision agriculture, and the source code and data are available at https://github.com/yangxin6/3D-PACA-Network.git.
Why it matches plant phenotyping methods3D点群による個体分割と形質抽出ネットワークを開発し、複数データセットで性能評価・検証しているため、植物フェノタイピング手法が中心である。
abstractwe propose a novel framework named Paired-Attention Central Axis Aggregation Network (PACANet) for 3D point cloud-based plant segmentation.
Yield and its components are the important traits for plant breeders to select the best genotypes in the breeding programs. However, traditional measurements of these traits across genotypes and environments are labor-intensive and time-consuming, as hundreds or even thousands of plots need to be estimated. A yield trial was carried out using seven sugarcane cultivars planted in a randomized complete block design with four replications for two ratoon crops to estimate sugarcane yield and its components using unoccupied aerial systems (UAS)-based high throughput phenotyping (HTP) and to compare the traditional method with UAS-based yield components in discriminating ability to assess sugarcane yield via a path coefficient analysis. UAS platforms mounted with sensors were flown over the trial. The result shows that UAS-derived plant height (PH) showed a strong relationship with the ground measured PH (R² = 0.89, RMSE = 0.15 m). Likewise, an accurate millable stalk height (MSH) estimation, using UAS-derived PH as a predictor, was observed (R² = 0.54, RMSE = 0.15 m). Canopy height model (CHM)-derived canopy cover (CC) appeared to be a promising feature to indirectly select or to predict for stalk number (SN) (R² = 0.69, RMSE = 10,975 stalks ha⁻¹). Based on a path coefficient analysis, UAS-based yield components performed equally to or slightly underperformed the traditional method. Traditionally, SN was the largest contributor to cane yield. Similarly, CC and CHM were the important components for UAS-based yield components. Additionally, the yield prediction model using UAS-derived canopy features with five cross validation schemes (CVs) revealed that model accuracy increased as association between predictor variables with a responding variable increased. The present study shows that random forest outperformed (higher r and lower RMSE) the linear regression models (stepwise, lasso, and ridge) in all CVs. The linear regressions were off when they were used to predict the performance of cultivars in untested crop/environments (CVs2 and CVs5), while a higher accuracy was observed when using random forest in those CVs. More importantly, the accuracy of all models reduced when they were tested in untested crop/environments (CVs2 and CVs5), indicating the challenge of using a prediction model applied to new environments.
Why it matches plant phenotyping methodsUASベースHTPでサトウキビの草高・群落被覆・茎数・収量構成要素を推定し、地上測定との検証、モデル比較、交差検証を行っており、表現型取得・推定手法が研究の中心である。
abstractUAS-derived plant height (PH) showed a strong relationship with the ground measured PH (R² = 0.89, RMSE = 0.15 m).
PeachAerial / UAVField / plotFlowerClassificationSegmentationGrowth / development / phenology
A timely and accurate assessment of flowering characteristics is vital for tracking floral phenology in agricultural management and peach breeding. In this study, an unmanned aerial vehicle (UAV) equipped with a high-resolution camera was integrated with deep learning techniques to monitor peach flowering across multiple varieties. An instance segmentation model, PeFloSEG, was proposed for the accurate detection of peach flowers and buds. Based on the YOLOv5-seg framework, PeFloSEG integrates an enhanced detection head for improved feature representation and a modified loss function—Focal Efficient Intersection over Union (Focal-EIoU)—to optimize bounding box regression. To boost model efficiency, a network slimming algorithm was applied, significantly reducing model size while maintaining high accuracy. PeFloSEG achieved strong results, with mean average precision (mAP@0.5) scores of 0.876 for detection and 0.825 for segmentation, outperforming state-of-the-art algorithms by 0.5 %–24.9 % in detection and 5.7 %–28.1 % in segmentation. Three flowering indices derived from PeFloSEG outputs—flowering intensity (FI) indices (FI1 and FI2) and single-tree flowering ratio (SFR)—were evaluated. Linear correlation analysis revealed strong relationships between these indices and ground truth values, with R² values of 0.964 (FI1), 0.961 (FI2), and 0.986 (SFR). These indices were further used to assess flowering dynamics over time and to distinguish phenological stages, achieving an overall classification accuracy of 91.7 %. They also enabled effective variety classification, facilitating the exploration of flowering characteristics across different peach varieties. Overall, the proposed approach offers a scalable and efficient solution for high-throughput phenotyping and provides valuable tools for peach breeding and germplasm resource evaluation.
Why it matches plant phenotyping methodsUAV画像とインスタンスセグメンテーションモデルを開発し、花数・開花強度・単木開花率などの植物表現型を定量化することが研究の中心であるため。
abstractAn instance segmentation model, PeFloSEG, was proposed for the accurate detection of peach flowers and buds.
Leaves play a fundamental role in the plant body by performing photosynthesis. Their morphological characteristics, leaf area and other surface parameters, can help to explain various processes, such as climate change, ecological relationships, and agricultural productivity. However, most existing methods for measuring leaf surface dimensions are expensive and often complicated. Additionally, several methods employ destructive approaches, preventing the monitoring of plant growth. In this work, we introduce a new deep neural network to estimate bean’s leaf area from images containing a salient leaf and a marker. Our method is based on the DeepLabv3+ architecture for semantic segmentation and comprises one encoder and two decoders, which estimate the image segmentation and the pixel areas of the objects of interest. An extensive quantitative and qualitative analysis was conducted with the model’s predictions on 3374 images of 300 different leaves. Results indicate that the trained model is capable of estimating the leaf and marker areas with only one input image.
Why it matches plant phenotyping methods画像から葉面積を非破壊推定する深層学習手法を開発し、多数画像で評価しており、植物表現型の取得・抽出が研究の中心である。
titleNon-destructive leaf area estimation based on a semantic segmentation deep neural network
Wheat is one of the three primary staple crops globally, with the senescence of its leaves having a direct effect on yield. However, conventional senescence evaluation methods are mainly based on visual scoring, which are subjective, time-consuming, and hamper the investigation of mechanisms between senescence process and yield formation. High-throughput image-based plant phenotyping techniques offer a promising approach. However, extracting senescence-related semantic information from images presents challenges, including blurred edge segmentation, inadequate characterization of senescence features, and interference from complex field environments. Therefore, this study proposes a dual-branch image senescence segmentation model (SenNet), which integrates edge priors and local–global attention mechanisms, including local–global hierarchical attention mechanisms, gated convolution, and positional encoding modules. First, a wheat senescence dynamics image dataset (19530 images) was constructed, comprising 509 wheat varieties from a two-year and two-replicate field experiments. Then, the SenNet model achieved senescence image segmentation for various wheat varieties, enabling senescence dynamics analysis and high-yielding variety screening. The results showed that: 1) The mean Intersection over Union (mIoU) of the SenNet model was 95.41 %, which represented a 4.01 % improvement over the average mIoU of seven state-of-the-art models. 2) The contributions of the local–global hierarchical attention mechanism, gated convolution, and positional encoding module to the accuracy improvement of SenNet were 3.15 %, 1.62 %, and 1.03 %, respectively. 3) SenNet can be transferred across years and locations. The mIoU accuracy of the SenNet across locations is 96.01 %. Furthermore, the model trained in 2023 can be transferred to 2022 and 2024, achieving mIoU accuracies of 93.75 % and 93.27 %. 4) High-yielding varieties typically experience a later onset of senescence and faster senescence in later stages. Based on the senescence law, this study further constructed new dynamic traits of senescence (e.g., AreaUnderCurve). Leveraging the random forest-based yield prediction (R² = 0.68) from the dynamic traits, high-yielding varieties were screened with an average precision, recall, F1 score, and accuracy of 81 %, 79 %, 80 %, and 87 %, respectively. This study provides an efficient method for monitoring senescence dynamics and predicting yield, offering new insights into the screening of high-yielding varieties.
Why it matches plant phenotyping methodsコムギ葉の老化状態を画像から分割・定量するSenNetを開発し、データセット構築、モデル比較、地域・年次移植性を検証しているため、植物表現型取得手法が中心である。
abstractTherefore, this study proposes a dual-branch image senescence segmentation model (SenNet)
Cymbidium hybrids has the advantages of strong growth, large number of flowers, and high market share. However, most orchid cultivation has problems such as over-reliance on artificial experience cultivation, lack of scientific basis for water and fertilizer application, and low degree of precision and intelligent cultivation. Monitoring the levels of N, P, and K in plant leaves facilitates real-time assessment of nutritional status while guiding optimization fertilization, thereby enhancing fertilizer efficiency and plant growth quality. Nevertheless, conventional nutrient detection methods are invasive, time-consuming, and expertise-dependent, necessitating advanced non-destructive alternatives. This study explores hyperspectral sensing technology for non-destructive monitoring of N, P, and K in hybrid orchid under greenhouse conditions. Three water-fertilization regimes were applied at strategic growth stages of hybrid orchids, with spectral data collected across four seasons using a spectrometer and a hyperspectral imaging device. Through the analysis of 720 combinations of ten spectral preprocessing methods and four modeling approaches, an optimal algorithm combination was identified. This optimal combination was integrated with the Stable Competitive Adaptive Reweighted Sampling algorithm to select spectral feature bands. Predictive models were then constructed to estimate the contents of N, P, and K in the orchid leaves. The results showed the utilization of Stable Competitive Adaptive Reweighted Sampling method, between 14.2% and 25.0% of the feature wavelengths significantly reduced the root mean square error of the models. By incorporating these methodologies, the predictive model for leaf N, P, and K contents utilizing the hyperspectral imaging device achieved determination coefficients (R²) of 0.7395, 0.7213, and 0.7055, respectively, on the test sets. For the spectrometer-based models predicting leaf N, P, and K contents, the corresponding R² values for the test sets reached 0.8708, 0.8747, and 0.8557. These results validate hyperspectral sensing as a robust tool for non-destructive nutrient monitoring. The findings of this study provide a theoretical basis and effective methodology for the accurate and non-destructive detection of nutrient content in orchid leaves through hyperspectral sensing technology.
Why it matches plant phenotyping methodsランの葉内N・P・K含量という植物状態を、ハイパースペクトルセンシングで非破壊推定するモデルを開発・検証しており、表現型取得手法が中心的である。
abstractThis study explores hyperspectral sensing technology for non-destructive monitoring of N, P, and K in hybrid orchid under greenhouse conditions.
To address the challenges of multimodal data fusion, low deployment efficiency, and inadequate recognition robustness in complex environments for fruit tree disease segmentation and severity classification, a multimodal parallel transformer-based framework was proposed for apple disease recognition and grading. This method integrates image data with multi-dimensional environmental sensor information. An image segmentation preprocessing module was incorporated to enhance lesion region representation, while a cross-scale attention mechanism and a frame-wise diffusion module were introduced to improve robustness under challenging backgrounds. Additionally, pruning, quantization, and knowledge distillation techniques were employed to enable lightweight deployment. Experimental results demonstrated that the full model achieved outstanding performance on apple disease recognition tasks, reaching a precision of 0.98, recall of 0.93, F1-score of 0.95, and accuracy of 0.96, surpassing several state-of-the-art methods including Mask R-CNN, SegFormer, and Swin Transformer. After compression, the model size was reduced to 76.4 MB, and computational complexity decreased to 6.1 G, enabling real-time inference speeds of 25.2 FPS and 39.6 FPS on Jetson Xavier and Orin platforms, respectively. Ablation studies confirmed the performance contributions of the segmentation preprocessing, sensor fusion, and diffusion modules, demonstrating the potential of the proposed framework for deployment in resource-constrained agricultural scenarios.
Why it matches plant phenotyping methodsリンゴ病斑の画像セグメンテーションと重症度分類を中核とする軽量マルチモーダル手法を開発し、性能比較・アブレーション・実装性能評価まで行っているため、植物フェノタイピング手法として含める。
abstracta multimodal parallel transformer-based framework was proposed for apple disease recognition and grading
Accurate and efficient crop height information retrieval is crucial for applications such as farmland management, growth monitoring, yield estimation, and pest monitoring. Polarimetric Synthetic Aperture Radar (PolSAR) is known for its high sensitivity to the shape, structure, and dielectric constant of vegetation, presenting great potential for crop height retrieval. In this study, we compare the performance of three machine learning algorithms, Random Forest Regression (RFR), Bagging Decision Tree (BAGTREE), and Extreme Gradient Boosting (XGBoost), in the retrieval of crop height from PolSAR data. Using a comprehensive approach, we constructed a set of 32 polarimetric features as the initial input for the model. Subsequently, feature selection is employed to generate a subset aimed at reducing redundancy and improving the final estimation accuracy. Multi-temporal C-band PolSAR RADARSAT-2 data collected over three distinct agricultural types (corn, wheat, and soybean) in Canada are chosen for this study. The results indicate that the optimal average Root Mean Square Error (RMSE) for height retrieval in corn, wheat, and soybean throughout their growth cycles is 43.69 cm, 10.78 cm, and 20.92 cm, respectively. Among the three algorithms, RFR consistently demonstrates stable retrieval performance, and the polarimetric decomposition parameters exhibit the highest sensitivity to crop height. This study offers a valuable technical reference for SAR-based crop height retrieval and remote sensing-based crop growth monitoring without interferometry.
Why it matches plant phenotyping methodsPolSARセンサーデータと機械学習により作物高を推定し、複数アルゴリズムを比較・検証することが研究の中心であるため、植物形質計測手法として含める。
titleCrop height retrieval from polarimetric SAR data using machine learning: A comparative and validation study
Plant diseases establish a serious risk to worldwide food safety and agronomic sustainability, leading to substantial losses in harvest and quality. Precise and appropriate experience of these viruses is necessary for employing effective control procedures, diminishing economic victims and confirming food accessibility. However, current detection methods are often hindered by limitations such as insufficient generalizability across various crops, difficulty in handling complex, and noisy backgrounds, and suboptimal performance when addressing diverse and overlapping disease symptoms. This research is driven by the persistent necessity to tackle these issues and delivers a novel approach to improve disease identification and categorization performance. For this need, propose the Efficient Capsule convolutional Shuffle Attention Network (ECSAN), a comprehensive framework specifically designed to classify diseases across five diverse crops: Apple Leaves, Cassava Leaves, Hibiscus, Hyacinth Bean, and Okra Leaves. The framework integrates a fast gradient-domain weighted guided image filter to denoise and improve image superiority and segmentation is done over a dense swin transformer combined with Unet. Statistical features are extracted using general kernel joint non-negative matrix factorization, and the classification progression is optimized through the Enhanced Osprey Optimization Algorithm (EOOA). Implementation outcomes on five standard datasets prove the superiority of ECSAN, achieving 99.9 % accuracy, 99.98 % recall, 99.97 % precision, 99.98 % F1-score, and 99.99 % specificity. Additionally, ECSAN significantly reduces execution time compared to existing methods, emphasizing its efficiency and applicability in agricultural scenarios. This research underscores the urgency of addressing current detection challenges and establishes a robust foundation for advancements in plant pathology.
Why it matches plant phenotyping methods植物葉の病害状態を画像から抽出・分類する新規深層学習フレームワークを開発し、複数作物・標準データセットで性能評価しているため、植物フェノタイピング手法が中心である。
abstractdelivers a novel approach to improve disease identification and categorization performance
Typhoons are a primary cause of maize lodging, significantly reducing crop yield and resilience. Rapid and accurate assessment of lodging severity is essential for processing agricultural decision and implementing effective cropland management strategies. However, the crop lodging monitoring methods based on remote sensing require a large number of ground survey samples, which are time-consuming and labor-intensive, limiting their applicability for large-scale assessments during typhoon events. To address this challenge, this study proposed a Two-Step Augmentation Strategy (TSAS) that integrates Sentinel-2 and Sentinel-1 satellite data with limited field samples to automatically generate representative samples for different lodging severity. The TSAS framework includes two steps: first step, we use eXtreme Gradient Boosting (XGBoost) to classify random points in overlapping regions (RegionOₗ, referring to areas covered by both Sentinel-1 and Sentinel-2 images) based on lodging-sensitive optical and radar features identified through correlation analysis and J-M distance, generating Sample Set L. Second step, we retrain the XGBoost model with radar features from Sample Set L to classify random points in non-overlapping regions (RegionNₒₗ, referring to areas covered only by Sentinel-1 images and not by Sentinel-2 images). Finally, a purification strategy is applied to remove erroneous samples and improve accuracy. This approach was tested in Jilin Province, significantly affected by a severe typhoon in 2020. Results show that TSAS effectively generates reliable lodging samples, achieving high spectral correlation similarity (mean of SCS > 0.72) and low Euclidean distance (mean of ED < 0.75) compared to field survey samples. These samples were applied to three classifiers, with Support Vector Machine (SVM) achieving the highest accuracy (OA = 82.36 %, F1 = 0.81) compared to Random Forest (RF) and Minimum Distance (MD) classification methods. The results of this study show that TSAS is able to efficiently and accurately generate high-accuracy maize lodging samples on a regional scale with limited field surveys, significantly improves the efficiency of large-scale agricultural disaster assessment and provides valuable support for disaster management and agricultural planning.
Why it matches plant phenotyping methodsSentinel-1/2データとXGBoostを用いて、トウモロコシの倒伏重症度サンプルを限られた現地調査から自動生成するTSAS手法が研究の中心であり、植物状態の推定と精度評価を実施している。
abstractthis study proposed a Two-Step Augmentation Strategy (TSAS) that integrates Sentinel-2 and Sentinel-1 satellite data with limited field samples to automatically generate representative samples for different lodging severity.
Plants are subjected to a plethora of biotic stresses caused by various pathogens; among them, fungal pathogens represent the most destructive ones. In order to preserve the health status of plants, especially under the influence of climate change, the need to develop new sustainable, inexpensive, in-field and non-destructive diagnostic methods for plant pathogens is of great importance. In this direction, spectroscopic and molecular methods have made progress, while others, such as electrical diagnostic methods are still in the early stages of development. In this work, electrical signals in tomato plants infected with the fungal pathogen Oidium neolycopersici, the causative agent of powdery mildew, were measured. Differences in electrical responses were observed between healthy and infected plants during the entire monitoring period, and infected plants showed overall lower values of the electrical potential in comparison with healthy plants. Measurement of electrical potential allowed the successful differentiation between infected and healthy plants before the onset of symptoms (3.2 days in advance). A significant difference in electrical signals was obtained not only between infected and healthy plants, but also concerning the growing substrate: A stronger electrical potential was measured in plants grown in the peat-based substrate compared to those cultivated in the water substrate allowing a 97.5% discrimination. Based on the results of this study, measurements of electrical signals may become the basis for an alternative non-destructive diagnosis of tomato powdery mildew and other plant diseases. With the possibility of directly applying the technique in the field followed by remote monitoring of electrical signals, it may become useful for supporting timely disease control.
Why it matches plant phenotyping methodsトマトの感染状態を電気信号から非破壊的に識別する診断手法を測定・評価しており、植物病害状態のフェノタイピング手法が中心です。
abstractelectrical diagnostic methods are still in the early stages of development
Variable-rate spraying technology based on plant canopy volume is widely applied during precision agricultural practices. However, field observations indicate that larger canopies typically correlate with healthier trees, exhibiting reduced disease risk and severity. Therefore, the integration of canopy health information into variable-rate pesticide application is imperative. This study developed a variable-rate spraying system that addresses the critical limitations of conventional variable-rate spraying systems by employing a multisensory fusion methodology, which integrates real-time spot detection with plant canopy volumetry. Dual-modality sensing system achieves concurrent canopy volume measurement (4.34% error) and spot detection (82.6% accuracy), enabling adaptive pesticide dosage optimization through integrated decision-making that synergizes disease severity with canopy volumetry. Experimental tests demonstrated a 74.0% reduction in chemical usage compared to conventional spraying, while maintaining the required deposition parameters for low-severity diseases, and a 42.7% reduction for high-severity diseases. When targeting pear rust disease, the system attained 86.53% control efficacy with 66.4% and 22.2% dosage savings over traditional and volume-only variable-rate spraying approaches, respectively. The newly developed variable-rate spraying system can perform as a precise variable-rate spraying system using fused information on fruit-tree canopy volumes and disease, which significantly reduces the use of pesticides without affecting the control effect, further enabling a reduction in pesticide quantity and an increase in efficiency.
Why it matches plant phenotyping methods植物キャノピー体積と病害スポットを同時測定するマルチセンサ計測システムを開発・検証しており、植物の形態・病害状態の取得が散布最適化の中心的技術貢献である。
abstractThis study developed a variable-rate spraying system that addresses the critical limitations of conventional variable-rate spraying systems by employing a multisensory fusion methodology, which integrates real-time spot detection with plant canopy volumetry.
Yield forecasting is crucial for growers, enabling efficient resource management and informed decision-making. Such decisions impact storage, product processing, and logistics, leading to increased productivity and cost savings. However, this heavily relies on accurate yield forecasts. This work addresses such a need by presenting the development and testing of a reliable method for yield forecasting. The proposed methodology combines high-resolution object detection with a multi-variate input forecasting model that accurately computes the yield for incoming harvests. The forecasting approach incorporates a physically-constrained model based on a Long Short-Term Memory (LSTM) network. This model dynamically applies weights to the time-series data composed of counts for the phenological stages: flower, green, small white, large white, pink, and red (ripe fruit). These counts are obtained from detections made by a YOLOv10s, achieving an mAP@50 of 0.74 for all classes. As a result, the forecasting model's capacity to interpret input data is enhanced, translating it into a valid ripe count forecast. To validate the proposed approach, the forecasting model was trained and evaluated using (a) untreated count sequences and (b) weighted count sequences. The results indicate that phenologically-weighted input sequences outperform untreated sequences, with the following evaluation metrics: R² = 0.74, Root Mean Square Error (RMSE) = 12.67, Mean Absolute Error (MAE) = 10.95, and Mean Absolute Percentage Error (MAPE) = 39.4, improving 15%, 19.26%, 17.13%, and 11.3%, respectively.
Why it matches plant phenotyping methods画像ベースのYOLO検出でイチゴの生育段階・果実数を抽出し、収量予測へ利用する方法を開発・検証しており、植物表現型の取得と解析が中心的な貢献である。
abstractThis work addresses such a need by presenting the development and testing of a reliable method for yield forecasting.
Tea diseases cause significant economic losses to the tea industry every year, and thus developing a rapid and accurate tea disease detector is of great significance for assisting farmers in preventing diseases and increasing their income. Therefore, this paper proposes a lightweight and efficient detector called TDDet to quickly and accurately detect tea diseases. TDDet is mainly composed of two key innovations: feature extraction and feature aggregation. For feature extraction, we use lightweight depthwise separable convolution to reduce the computational load and enhance the ability to extract key local features in images of tea diseases. In addition, attention mechanisms including channel-, spatial-, and self-attentions, are employed to enable the model to focus on the most important parts of tea diseases, thereby improving the performance of the model. For feature aggregation, we propose a novel Cross-scale Feature Fusion (CFF) module to focus on tea disease areas, boosting the model’s sensitivity to feature details. Based on CFF, TDDet repeatedly fuses multiscale features of different levels in a top-down and bottom-up manner, enhancing feature representation capability. Besides, a lightweight and efficient upsampling module, called Dysample, is used to reduce computational costs and improve model performance by dynamically adjusting the sampling rate of feature maps. Experimental results demonstrate that TDDet with fewer parameters outperforms other state-of-the-art object detection models, enabling fast and accurate identification of tea diseases. Our code and dataset are available at https://github.com/hpguo1982/TDDet.
Why it matches plant phenotyping methods茶葉の病害画像から病害を検出する軽量モデルを開発し、性能比較も行っており、植物病害状態の取得手法が研究の中心である。
abstractthis paper proposes a lightweight and efficient detector called TDDet to quickly and accurately detect tea diseases
Traditionally, destructive analysis of the internal chemical components of plants is necessary to assess their overall health. This study proposes a novel approach to non-destructively estimate and classify organic and inorganic components associated with the general health of kimchi cabbage by integrating spectral imaging and system dynamic modeling techniques. Existing vegetation indices rely on constant values, which limits their ability to classify plants with similar measured constant values but different internal component contents. This problem is addressed by systematically approaching the vegetation indices and extracting and using intrinsic steady-state value and response-velocity parameters. It uses the principle that healthy plant pixel data have low-pass filter characteristics, and reflectance data from stressed or damaged plant areas exhibit high-pass filter characteristics. The proposed dynamic model identifies the relationship between red and near-infrared wavelength reflectances as time series data, and this framework can transform constant-based vegetation indices into dynamic system-based models. The mechanistic model improved accuracy by 33 % for Lutein, 10.5 % for beta-carotene, 8 % for Chlorophyll-a, and 15.8 % for Chlorophyll-b compared to the existing method. In addition, differences in calcium content between treatment groups, which are difficult to resolve using traditional vegetation indices, were identified. The dynamic model provides a solution for simultaneous, non-destructive analysis of pigments and calcium components, which traditionally relied on destructive testing. This scalable and efficient technology has great potential to bridge the gap between precision agriculture and conventional agriculture and contribute to sustainable agricultural realization.
Why it matches plant phenotyping methodsスペクトルイメージングと動的モデルを用いて、植物の健康状態や色素・カルシウム含量を非破壊推定する手法の開発が中心であり、植物フェノタイピング手法に該当する。
abstractThis study proposes a novel approach to non-destructively estimate and classify organic and inorganic components associated with the general health of kimchi cabbage by integrating spectral imaging and system dynamic modeling techniques.
An early and rapid detection of nitrogen (N) stress in field crops is crucial to mitigating nutrient deficiency and achieving sustainable crop yield. Although numerous methods and equipment have been developed to monitor crop N stress and fertilizer application thereof, many of these technologies face significant limitations in terms of costs, accuracy, integration, etc. This study reports the development of a Variable Rate fertilizer Application (VRA) system assisted by Deep Learning (DL) model deployed embedded system to enable rapid RGB image-based detection of nitrogen stress in wheat crop and subsequent application of N fertilizer. AlexNet DL model resulted in precision, recall, and F1-score as 0.977, 0.973, and 0.973, respectively; for classifying N stress into three classes. The developed VRA could operate in sync with embedded system at an operational speed of 0.4 m/s with a field capacity of 0.32 ha/h in a 26 DAS wheat crop. The effectivity of the VRA was evaluated by vegetation indices (ExG, RGRI, VARI and NGRDI) with drone assisted RGB images before and after VRA operation; there was a consistent difference in before and after average index values for ExG (0.2046 and 0.2917) and VARI (0.1478 and 0.2454). These results are indicative of the uniformity of operation by VRA throughout the field. The average percentage N fertilizer saving under VRA as compared to traditional technique was 37.53 % with an insignificant (p < 0.05) difference in yield. This study delivers a real-time effective technique for precise classification of N stress and its real-time mechanized management in wheat crop.
Why it matches plant phenotyping methods小麦の窒素ストレスという植物状態をRGB画像と深層学習でリアルタイム分類する手法および搭載VRAシステムの開発が研究の中心であり、実地性能も評価している。
abstractThis study reports the development of a Variable Rate fertilizer Application (VRA) system assisted by Deep Learning (DL) model deployed embedded system to enable rapid RGB image-based detection of nitrogen stress in wheat crop
Nitrogen, as a vital element for plant growth and development, significantly influences crop yields. Nitrogen deficiency severely impairs crop growth, while excess nitrogen harms the environment. To address this, there is an urgent need for rapid and on-site methods to assess the physiological status of crops under nitrogen stress. In this study, we utilized Raman spectroscopy, a non-destructive and rapid analytical technique, to evaluate the physiological status of wheat plants subjected to various nitrogen treatments. These treatments included optimal, low, excessive and zero nitrogen application. By leveraging Raman spectroscopy’s ability to identify characteristic peaks of metabolites in plant leaves and quantify them based on peak intensity, we analyzed the levels of carotenoids, chlorophylls, cellulose, lignin, and aliphatic components. Our results revealed significant differences in metabolite peak intensity under different nitrogen treatments. Optimal nitrogen application promoted the accumulation of metabolites, while nitrogen deficiency led to a marked decrease in photosynthetic pigments and structural components. Excessive nitrogen caused a reduction in lignin and cellulose. To diagnose nitrogen stress, we developed classification models that accurately distinguished between healthy and nitrogen-stressed plants, achieving a training set accuracy of 99 %, a 5-fold cross-validation accuracy of 92 %, and a prediction set accuracy of 93 %. Furthermore, we differentiated wheat plants with varying degrees of nitrogen deficiency, achieving a maximum accuracy of 78 %. When considering both nitrogen deficiency and excess, the maximum accuracy reached 58 %. This study provides a fast, accurate, and non-destructive analytical method for analyzing and diagnosing nitrogen stress in field wheat based on Raman spectroscopy. Future research aims to extend this approach to the diagnosis of nitrogen stress in other crops and to explore its applications in nitrogen fertilization management.
Why it matches plant phenotyping methodsラマン分光法を用いて圃場コムギの窒素ストレスという植物生理状態を非破壊・迅速に評価し、分類モデルの精度検証まで行っており、表現型取得手法が研究の中心である。
titleRaman spectroscopy as a non-destructive and rapid method
Sustainable and low-nicotine production of tobacco requires rapid and accurate on-site assessment of the leaf nitrogen (N) status. This issue can be supported by fluorescence-based sensors, which are promising tools for precision N management. We then aimed to 1) evaluate the suitability of the Multiplex® fluorescence sensor (Mx) to predict, at an early stage, the final nicotine content of tobacco leaves; 2) develop a model for in-season tobacco foliar N estimation using the Partial Least Square (PLS) multivariate regression technique; and finally, 3) test the effectiveness of a Mx map-based Variable Rate Nitrogen Fertilization (VRNF) in reducing the spatial variability in leaf Nitrogen Balance Index (NBI), that is the N status, of a commercial field of Virginia Bright tobacco. The NBI measured by the Mx about two months after transplanting was found to linearly relate to the nicotine content measured after curing (R² = 0.72, P < 0.001) over a nicotine range of 0.25 – 4.12 %. NBI, defined as the ratio between the leaf chlorophyll (SFRR) and Flavonoids (FLAV) indices better related to nicotine than the single SFRR and FLAV indices (R² = 0.47, P < 0.001 and R² = 0.52, P < 0.001, respectively. Furthermore, the NBI estimated the actual leaf N content before flowering better (R² = 0.33) than single SFRR and FLAV indices (R² = 0.28), over a range of 21 – 37.6 mgg⁻¹. Leaf fluorescence sensor indices were thus combined with growth stages and weather variables across diverse varieties and sites. The resulting PLS model successfully predicted leaf N (R² = 0.72, RMSEP = 2.73 mgg⁻¹ and relMAE = 7.75 %) over a range of 20.6–28.0 mgg⁻¹. The most significant variables, primarily related to solar radiation, were identified for a robust general model development. Finally, the spatial pattern of the NBI was mapped over a 2.04 ha commercial plot of the ITB 6118 variety, and used to produce a three-zone prescription map. Two weeks after the intervention of VRNF based on the defined prescription map, the overall NBI variability had dropped from 23.5 % coefficient of variation (CV) to 7.9 % CV. Our results show the feasibility of using the Mx sensor for precision fertilization of Virginia Bright tobacco and highlight its potential to support future developments aimed at more sustainable production of plants with reduced nicotine content.
Why it matches plant phenotyping methods蛍光センサーによる葉のニコチン含量・窒素状態の推定モデルを開発・検証し、NBIマッピングと可変施肥へ応用しており、植物形質の取得・推定手法が中心である。
abstractevaluate the suitability of the Multiplex® fluorescence sensor (Mx) to predict, at an early stage, the final nicotine content of tobacco leaves
Accurate prediction of individual apple tree yields during the preharvest stage is essential for precision orchard management and market planning. However, systematic studies focusing on apple yield estimation are scarce. To address this gap, this study targets Fuji apples in the Aksu region of Xinjiang. Multiple images were captured via a UAV during four key growth stages: flowering, fruit formation, fruit expansion, and ripening. On the basis of the extracted vegetation indices, we first developed yield estimation models using random forest (RF), support vector regression (SVR), partial least squares regression (PLS), and ridge regression (RR) methods. We subsequently combined these four models to construct a stacking ensemble learning (SEL) model. To further increase the accuracy of apple yield estimation, we refined the growth stage stacking method and developed a new model, the growth stage stacking ensemble (GSSE). This model maximises the use of spectral information from multiple apple growth stages by employing various machine learning algorithms and integrating multistage spectral data to improve yield estimation accuracy. The results indicate that the optimal period for yield estimation occurs during the fruit expansion stage, with the support vector regression (SVR) model achieving the best performance (R² = 0.654, RMSE = 5.307 kg). Compared with individual machine learning models, the SEL approach enhances yield estimation accuracy, reaching a maximum R² of 0.686 and an RMSE of 5.058 kg. Furthermore, GSSE significantly enhanced accuracy compared with the single-growth stage estimation models and SEL, with the combination of fruit expansion and fruit ripening stages yielding the best results, with an R² of 0.759 and an RMSE of 4.431 kg, with the fruit expansion stage contributing the most. This study is the first to apply the GSSE to apple yield estimation, offering novel insights for UAV-based apple yield estimation.
Why it matches plant phenotyping methodsUAV画像と多時期スペクトル情報から個体別リンゴ収量を推定するモデルを開発・比較しており、植物形質の取得・推定手法が研究の中心である。
abstractMultiple images were captured via a UAV during four key growth stages: flowering, fruit formation, fruit expansion, and ripening.
Remote sensing via unmanned aerial vehicle (UAV) could provide critical data support for estimating the real-time growth status of crops. In this study, the vegetation indexes (VI) and texture features (T) from multispectral images were extracted, and the entropy method was used to construct a comprehensive growth index (CGI) which ultimately reflected growth of winter wheat. Later on, the predictive models of winter wheat growth were established and evaluated by using the machine learning modeling methods of BPNN, RF and SVM with the different combinations of spectral and texture features. The results revealed that the correlation of CGI was improved compared with other single growth indicators, and it also reached a significant correlation level (r > 0.6) with the texture features based on the red edge band. For different input variables, the CGI estimation accuracy for most models based on the combination VI and T were superior than that of VI or T alone with the mean R² = 0.858; while the average values of R² of the models based on VI and T alone were 0.825 and 0.774 respectively. It also indicated that fusion of the spectral and texture features improved predictive performance of winter wheat growth. Among all the crop growth indicators, the CGI achieved the best performance as well by using the RF and VI + T variable (R² = 0.888, RMSE = 0.041, RPD = 2.989), which confirmed the application potential of RF machine learning method in estimating the winter wheat growth. Lastly, it also proved the feasibility of constructing comprehensive indicators to monitor wheat growth by entropy method as the fact that the estimated results of CGI models were also better than most of the single growth indicators with the mean R² = 0.819 for all the CGI models. This study is expected to offer both theoretical and practical references for monitoring the growth of winter wheat through UAV-based multispectral technology.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像から冬コムギの生育状態を推定する特徴抽出・指標構築・機械学習モデルを開発し、精度評価しており、植物表現型取得・推定が研究の中心である。
abstractthe vegetation indexes (VI) and texture features (T) from multispectral images were extracted, and the entropy method was used to construct a comprehensive growth index (CGI) which ultimately reflected growth of winter wheat.
PotatoRaman / spectroscopyPhysiological trait estimationWater status / transpiration
Near infrared spectroscopy (NIRS) has been widely used as a nondestructive testing technique and plays a crucial role in the quality inspection of agricultural products. However, the variability between different batches of samples hinders the application of commercial NIRS processes. Therefore, model transfer is usually performed on new samples to enhance the generalizability of the device. In this study, a new algorithm was developed based on the slope and bias correction algorithm (SBC) for model transfer between two different batches of samples, using potatoes of different origins as experimental samples for prediction models of potato quality, and based on this algorithm, model transfer between two different batches of samples was successfully implemented in a self-made portable non-destructive potato detection device. The results showed that the device developed based on the new algorithm gives good results for subsamples prediction. In the dry matter model, the correlation coefficient (R), root mean square error (RMSE) and relative standard deviation (RSD) of the new algorithm optimized compared with the traditional SBC algorithm were improved from 0.7843, 1.2080% and 6.59% to 0.8251, 1.1307 and 6.17%, respectively; in the starch model, the new algorithm optimized R, RMSE and RPD improved from 0.7971, 1.0023% and 7.43% to 0.8176, 0.9570% and 7.31%, respectively, compared with the traditional SBC algorithm. The transfer of NIR correction models for the dry matter and starch content of potatoes was basically achieved, which provided technical and theoretical support to enhance the model universality of convenient nondestructive detection devices.
Why it matches plant phenotyping methodsジャガイモの乾物・デンプン含量という植物器官形質を非破壊NIRで測定する携帯型デバイスと、バッチ間モデル移 transferアルゴリズムを開発・評価しており、表現型取得法が中心である。
abstracta new algorithm was developed based on the slope and bias correction algorithm (SBC) for model transfer between two different batches of samples
This paper presents a robotic system designed to replace manual labour in three prevalent activities related to blueberry management: soil sampling and analysis, weed spraying, and plant health status monitoring. A complex system for automating field activities in blueberry orchards that involves the use of ground and aerial robots, along with integrated route optimisation software, autonomous driving, image recognition based on AI, and others, was developed. The ground robot is made in a modular way, having three different add-on modules for the three distinct use cases it covers. A modular system guarantees the year-round utilisation of the robot across various growth stages of plants. This stands out as its major advantage, considering that many robotic systems are typically tailored to address a singular task. Robot modules utilise custom-made hardware for in-field sampling and real-time soil analysis, an industrial robotic arm with a custom-made system for spraying, and a plant health monitoring module consisting of two Plant-O-Meter™ optical devices capable of capturing sixteen optical vegetation indices in real-time. The solutions are tested and deployed in the real-world environment of the blueberry orchard. We have achieved 50–60 % savings on herbicides compared to blanket spraying, and the costs related to sub-optimal herbicide application are reduced up to 25 %. The costs of soil analysis have been significantly reduced to about 70 %.
Why it matches plant phenotyping methodsブルーベリー園向けロボット基盤の主要モジュールとして、植物健康状態を光学センサーでリアルタイム測定する方法を実環境で展開・評価しており、植物状態の取得が技術的構成の中心的要素の一つである。
abstracta plant health monitoring module consisting of two Plant-O-Meter™ optical devices capable of capturing sixteen optical vegetation indices in real-time
While tobacco plays a significant role in the global economy, research on regional tobacco growth simulation remains limited. This study integrates the WOFOST crop model with satellite remote sensing data, focusing on the data assimilation (DA) of leaf area index (LAI) to enhance the accuracy of regional tobacco growth simulations. Field survey data were used for model calibration, providing the foundation for the analysis. The performance of four 4-Dimensional Variational Assimilation algorithms (4DVAs)—Particle Swarm Optimization (PSO), Simulated Annealing (SA), Shuffled Complex Evolution-University of Arizona (SCE-UA), and Gray Wolf Optimization (GWO)—was compared with four sequential DA algorithms (SDAs)—Ensemble Kalman Filter (EnKF), Ensemble Variational (EnVar), Ensemble Square Root Filter (EnSRF), and Particle Filter (PF). The 4DVAs were developed by integrating constraint DA Algorithms (CDAs) into the 4D-Var framework, enhancing their capability to optimize model states over a time window. Additionally, the performance of their coupled DA algorithms was evaluated. The results indicated that the coupled of SA and PF (SA-PF) achieved the best performance in terms of model accuracy. Compared to field survey data for biomass, stem mass and leaf mass, our method achieved the coefficient of determination (R²) values of 0.89, 0.86, and 0.81, respectively, with normalized root mean square error (NRMSE) values of 0.12, 0.10, and 0.09. The SA-PF coupling algorithm also performs better than some new DA algorithms. This study provides a valuable reference for regional tobacco growth simulation and data assimilation, improving the accuracy and applicability of crop growth models.
Why it matches plant phenotyping methods衛星リモートセンシングのLAIを作物モデルへ同化し、バイオマス・茎重・葉重を推定するデータ同化手法を開発・比較・検証しており、植物形質推定が中心である。
abstractThis study integrates the WOFOST crop model with satellite remote sensing data, focusing on the data assimilation (DA) of leaf area index (LAI) to enhance the accuracy of regional tobacco growth simulations.
The precise segmentation of crop organs plays a crucial role in optimizing crop cultivation strategies and enhancing yield potential. This study proposes a novel deep learning network, CotSegNet, which enables precise and non-destructive segmentation of cotton organs facilitating the extraction of phenotypic characteristics. In CotSegNet, an improved attention mechanism known as CGLUConvFormer is designed. This mechanism significantly improves segmentation accuracy by emphasizing important features while diminishing redundant information. Furthermore, CotSegNet integrates the SegNext attention mechanism. This mechanism facilitates the efficient extraction and integration of multi-scale features, thereby significantly enhancing the ability of CotSegNet to comprehend and segment point cloud data. To address issues related to leaf adhesion and coplanarity that lead to over-segmentation problems, this study proposes an improved region-growing algorithm. This algorithm enhances the accuracy of leaf instance segmentation through the incorporation of distance constraints. In comparative experiments with five advanced deep learning networks (PointNet, PointNet++, DGCNN, SPoTr and CurveNet), CotSegNet demonstrated outstanding performance. Its Precision, Recall, F1-score, and IoU reached 95.06 %, 93.32 %, 94.61 %, and 89.80 %, respectively. The experimental results demonstrated that the proposed method effectively extracted the phenotypic parameters of stem height, leaf length, leaf width, and leaf area in cotton plants. These measurements exhibited a high degree of consistency with manual assessments, yielding determination coefficients of 0.947, 0.948, 0.955, and 0.961 for each parameter respectively. The corresponding root mean square errors were recorded as 0.852 cm, 0.492 cm, 0.551 cm, and 1.674 cm² respectively. The research findings demonstrate that this approach offers essential technical support for the collection and analysis of high throughput phenotyping data in field crops.
Why it matches plant phenotyping methods綿花器官の点群セグメンテーションと表現型形質抽出のためのCotSegNetおよび改良領域成長法を開発し、手動測定との検証も行っており、表現型取得が研究の中心である。
abstractThis study proposes a novel deep learning network, CotSegNet, which enables precise and non-destructive segmentation of cotton organs facilitating the extraction of phenotypic characteristics.
Drones have enabled large-scale breeding and cultivation experiments. However, extracting individual breeding plots from aerial images is a key prerequisite and urgent demand for extracting variety-level traits. The main difficulties in plot extraction include irregular rotation angles of the plots, ambiguous gaps both within and between plots, and variable color contrasts between the vegetation and the background. To solve these challenges, a novel oriented instance segmentation network (OSNet) is proposed by leveraging a global context transformer (GCT) and an oriented region proposal network (RPN). The performance was assessed using a well-labeled dataset with 960 plots of 160 wheat varieties across two years. Results show that OSNet achieved the AP@0.5 of 0.917, F1-score of 0.959, Accuracy of 0.966, IoU of 0.912, Recall of 0.934, and Plot-a of 0.999. OSNet outperformed five state-of-the-art (SOTA) networks with an average improvement of 3.08 %, 1.42 %, 1.19 %, 1.70 %, 1.79 %, and 0.04 % in AP@0.5, F1-score, Accuracy, IoU, Recall, and Plot-a, respectively. The sensitivity analysis proved that OSNet consistently achieved stable segmentation accuracy across different rotation angles and growth stages. The interpretability through ablation analysis showed that OSNet benefits from the oriented proposal and global information. Furthermore, OSNet can be transferred to new datasets with various years, crops, and data dimensions, supporting typical phenotyping tasks such as 2D wheat spike detection (r = 0.91) and 3D canopy height measurement (r = 0.89). The innovative methodology will be a fundamental tool for processing drone imagery, accelerating phenotypic trait extraction across various varieties and thereby expediting the breeding process.
Why it matches plant phenotyping methodsUAV画像から育種区画を抽出する新規インスタンスセグメンテーション手法を開発・検証しており、植物形質抽出への適用性能も評価しているため、方法が研究の中心である。
abstractTo solve these challenges, a novel oriented instance segmentation network (OSNet) is proposed by leveraging a global context transformer (GCT) and an oriented region proposal network (RPN).
Wheat diseases have severely threatened global food security, making prompt and accurate detection methods crucial for disease control. However, large-scale detection methods face challenges such as low accuracy, labor-intensive labeling processes, and limited applicability across different wheat diseases. This study proposes a novel method using an Unsupervised Domain Adaptation (UDA) for semantic segmentation, termed DATS-ASSFormer, to detect wheat rust, wheat scab, and wheat yellow dwarf from UAV-based multispectral images. The proposed method employs Domain Adaptation via Teacher-Student networks (DATS) to generate pseudo-labels for unlabeled data in the target domain, and Adaptive Semantic Segmentation Former (ASSFormer) to precisely segment diseased areas, thus minimizing the dependency on laborious manual labeling. To support the development and evaluation of the model, the Northwest A&F University-Wheat Disease Remote Sensing Dataset (NWAFU-WDRSD) was developed, encompassing 8,628 images of the three predominant wheat diseases. Extensive testing confirmed that the DATS-ASSFormer model outperformed existing models in six domain adaptation tasks, achieving average Oracle (supervised training within the target domain) and UDA mIoU scores of 85.80 % and 65.96 %, respectively. These results significantly enhance detection accuracy and robustness across various diseases in real-world agricultural settings. The efficacy of DATS-ASSFormer highlights its potential for practical applications in precision agriculture, offering a scalable and efficient solution for large-scale wheat disease detection and management. The project is accessible at https://github.com/YcZhangSing/DATS-ASSFormer.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像から小麦病害領域を抽出するセマンティックセグメンテーション手法を開発・評価し、専用データセットも構築しているため、植物病害表現型の取得が中心である。
abstractThis study proposes a novel method using an Unsupervised Domain Adaptation (UDA) for semantic segmentation, termed DATS-ASSFormer, to detect wheat rust, wheat scab, and wheat yellow dwarf from UAV-based multispectral images.
Fruit shape significantly impacts the quality and market value of chili peppers (Capsicum annuum). However, predicting their fruit shapes in F₁ hybrids remains challenging, often relying on skilled breeders. This study aimed to clarify the potential of elliptic Fourier descriptors (EFDs) to predict fruit shape of F₁ progeny in chili peppers based on parental data. Using images of 291 accessions (132 inbred and 159 F₁ from 20 parental inbreds), EFDs were extracted to reconstruct shape contours. The initial prediction method, PPₘᵢd, used midpoint EFDs of the parents, achieving accuracies comparable to genomic methods. To improve accuracy, a new method, PPδ, was developed. PPδ incorporates dominance effects observed in F₁ progeny, yielding significantly better predictions. Over 80% of F₁ accessions showed improved accuracy with PPδ, and the predicted contours aligned closely with real shapes. Cross-validation confirmed the reproducibility of PPδ predictions. These findings suggest that combining parental EFDs with dominance effect ratios enables accurate fruit shape predictions without genetic data. This is the first study demonstrating EFD applicability in F₁ hybrid breeding for fruit shape, offering a promising tool for developing innovative breeding techniques in chili peppers.
Why it matches plant phenotyping methods画像から抽出した楕円フーリエ記述子を用いてトウガラシ果実形状を予測する新手法PPδを開発し、交差検証で再現性を評価しており、果実形状フェノタイピング手法が研究の中心である。
abstractTo improve accuracy, a new method, PPδ, was developed.
Plant height (PH) of oilseed rape, as a crucial phenotypic indicator, provides essential data for seedling diagnosis and breeding selection when accurately monitored throughout the life cycle of individual plants. However, it is a challenge to obtain precise PH measurements as rapeseed leaves and other crops shade each other after flowering. In this study, a rail-based platform equipped with LiDAR and the BeiDou differential positioning system was designed and manufactured to autonomously collect time-series point cloud of oilseed rape populations in the field. The point cloud data of oilseed rape during the regreening stage was segmented using an improved fast Euclidean clustering algorithm, followed by extraction of the root collar region via an objective function. Centered on the identified root collar region, an adaptive plant envelope area (PEA) was generated based on the distance between adjacent plants to isolate individual rapeseed specimens. Within the PEA corresponding to each plant’s root collar region, the ground position during sowing and the canopy apex at distinct growth stages were precisely localized, enabling automated extraction of individual PH across the full life cycle. The coefficient of determination (R²) between the algorithm and the manual measurement results at 140, 150 and 165 days after sowing were 0.9742, 0.9667, and 0.9208, respectively. And Root Mean Square Error (RMSE) between the algorithm and the manual measurement results at 140, 150 and 165 days were 0.038, 0.043 and 0.061 m, respectively. These results confirm that integrating BeiDou positioning with 3D point cloud processing achieves high-precision phenotyping of crop height dynamics. Furthermore, PHs were applied to frost damage and lodging susceptibility analysis, which indicate that the growth rate of rapeseed slows down as the severity of frost damage increases, and plants that reach a height of approximately 1 m during the flowering stage are prone to lodging after rainfall. These results have the potential to provide guidance for frost damage assessment and variety selection in smart agriculture applications.
Why it matches plant phenotyping methodsLiDAR・BeiDou搭載プラットフォームと3D点群処理により、個体の生育全期間の草丈を自動抽出する手法を開発し、手測定と検証しているため、植物フェノタイピング手法が中心である。
abstracta rail-based platform equipped with LiDAR and the BeiDou differential positioning system was designed and manufactured to autonomously collect time-series point cloud of oilseed rape populations in the field.
Traditional methods for measuring pre-harvest loss, such as using quadrats, are labor-intensive and provide sparse data coverage. This study proposes an automated approach that leverages computer vision to replace and enhance the current method, using advanced imaging technologies and deep learning methodologies to detect and quantify pre-harvest losses in grain crops. Specifically, the methodology employs a camera mounted on the front snout of a ground vehicle, allowing continuous image capture along the crop rows. By automating image collection and analysis, this approach provides denser spatial coverage across the field, reduces errors associated with manual sampling and human judgment, and significantly accelerates the process compared to traditional quadrat sampling. In addition, this approach offers the potential to segregate different types of pre-harvest loss, such as natural shattering versus losses caused by mechanical disturbance, providing a level of granularity not achievable with conventional quadrat methods. By leveraging state-of-the-art object detection architectures the system is designed to handle the complex visual environment of the field floor, where grains may be obscured by crop residue, shadows, and similarly-colored objects such as stones. This capability represents a significant advancement over traditional methods, which cannot distinguish between these different loss sources. The images were annotated using the Segment Anything Model (SAM) to ensure consistency and accuracy across the dataset. Several state-of-the-art models were trained and evaluated on the collected data, including Mask RCNN, YOLOX, DETR, and a modified YOLOv8-p2. The modified YOLOv8-p2 model, which incorporated a p2 head to improve the detection of smaller objects, outperformed the others, yielding the highest Precision, Recall, and F1 scores on both the soybean (Precision = 0.727, Recall = 0.694, F1 = 0.710) and wheat (Precision = 0.709, Recall = 0.688, F1 = 0.698) datasets. Integrating the 850 nm NIR image channel did not produce a meaningful boost in performance, as evidenced by the soybean (Precision = 0.741, Recall = 0.689, F1 = 0.715) and wheat (Precision = 0.729, Recall = 0.690, F1 = 0.709) results. This research demonstrates that it is possible to integrate a vision system on the header of a combine and identify the initial shedding loss in the field. Future work will focus on refining the models further, exploring their applicability to other crop types, and integrating real-time processing and automation in data collection and analysis.
Why it matches plant phenotyping methods穀粒の収穫前損失という植物・作物状態を、車載カメラとコンピュータビジョンで自動検出・定量する手法を開発し、複数モデルで性能評価しており、表現型取得が研究の中心である。
abstractThis study proposes an automated approach that leverages computer vision to replace and enhance the current method, using advanced imaging technologies and deep learning methodologies to detect and quantify pre-harvest losses in grain crops.
Poplar trees are widely cultivated for their ecological and economic benefits. Studying the phenotypes of poplar seedlings can enable the selection of optimal cultivation methods to enhance yield and quality. UAV-based low-altitude remote sensing with optical sensors captures images and spectral data for such studies. However, deep learning in UAV plant phenotyping faces the challenge of requiring substantial time and effort to label image samples for model training. This paper aims to assess the efficiency of using Grounding DINO-SAM2 for zero-shot instance segmentation of individual poplar seedlings across multiple genotypes. An automatic program calculates image features from RGB and multispectral mask areas, including canopy projection, color, texture, and spectral reflectance, which are then used to establish a biomass estimation model based on two years of data. The study obtained the following results: (1) The Grounding DINO-SAM2 model was used to implement zero-labelled sample instance segmentation of 400 image data. After modifying the sample with incorrect target recognition quantity in less than 15 min, the total model took only 0.5 h, with a precision of 0.943, which greatly saved time and computing cost compared with mainstream fully-supervised segmentation models. (2) A poplar seedling biomass estimation model based on multimodal image features was established. After comparing and optimizing single-sensor and multi-sensor combined with different modelling algorithms, it was found that the CNN test set accuracy (R²) reached 0.823. This research provides a lightweight, cost-effective approach for plant image segmentation and feature extraction, promoting advances in intelligent management and monitoring for agriculture and forestry.
Why it matches plant phenotyping methodsUAV画像による個体セグメンテーションと特徴抽出を開発・評価し、ポプラ苗のバイオマスを推定する方法が研究の中心である。
abstractThis paper aims to assess the efficiency of using Grounding DINO-SAM2 for zero-shot instance segmentation of individual poplar seedlings across multiple genotypes.
Global climate change-induced environmental stress poses critical challenges to the stable development of economic crops such as citrus. High temperatures (HTs) at anthesis may cause poor pollination and excessive flower/fruit drop, seriously affecting fruit yield and quality. To comprehensively analyze the developmental dynamics and morphological responses of citrus to HT stress at anthesis, methods for precise whole-flower phenotypic extraction and stamen state classification were developed. A citrus flower automatic segmentation and phenotypic quantitative model (CF-ASPM) that combines the pre-trained Segment Anything Model (SAM) with a lightweight classification module was constructed to accurately identify and quantify key citrus flower structures. Phenotypic parameter extraction correlation coefficients were 0.90–0.98. A few-shot stamen classification method was also designed using a pre-segmentation strategy and differential features, and its classification accuracy was 96.39%. Experiments with Ehime mandarin were conducted to analyze dynamic citrus floral organ changes at different temperatures and the underlying physiological mechanisms. The results showed that citrus exhibits a distinct reproductive priority strategy under HTs. Floral organ growth is inhibited, blooming is accelerated, and an asynchronous compensation mechanism occurs between male and female organs. HTs accelerated flower aging and caused developmental imbalances in the ovary and nectar disc. This may lead to increased flower and fruit drop and altered fruit shape. This study revealed the reproductive priority strategy and growth imbalance of citrus floral organs under HTs using the CF-ASPM model. It provides important data for further exploring the molecular mechanisms and management strategies of HT stress.
Why it matches plant phenotyping methods柑橘花器官の自動セグメンテーション、形質抽出、雄蕊状態分類法を開発し、精度検証したことが研究の中心であるため。
abstractmethods for precise whole-flower phenotypic extraction and stamen state classification were developed.
Rice blast (RB) is a global fungal threat that occurs over multiple growth stages. The spatially explicit mapping of RB severity with remote sensing is crucial for precision crop protection. However, the interactions between spectral variations caused by the pathogen infection and phenological growth remain poorly understood in disease monitoring. This interference prevents the successful mapping of disease dynamics by introducing substantial errors in areas dominated by healthy plants. This study aimed to reveal the phenological influence on RB detection and to determine key spectral indicators for accurate classification for eliminating the interference of healthy plants on severity assessments. To achieve this goal, experimental data across growth stages were collected and comprised ground truth evaluations to derive the disease index (DI), and ground-based canopy reflectance and hyperspectral images acquired by an unpiloted aerial system (UAS). These datasets were used to examine the spectral responses to both i) RB infection, and ii) phenology, assessing the false positives obtained in the RB detection. Moreover, a two-step method was applied including the RB detection based on a novel feature selection method termed sequential importance selection (SIS) and the DI estimation using linear models based on rice blast indices (RIBIs). The results demonstrated that the spectral signatures in responses to phenology and RB infection were highly similar in the red and near-infrared regions, as well as in traditional vegetation indices (VIs) associated with plant traits. Such similarity yielded considerable false positive rates (FPR) in RB detection and pseudo-DI in healthy plots when applying individual VIs. In RB detection, the VIs selected by SIS (VISIS) achieved significantly higher overall accuracy (OA) and lower FPR than the best RIBI variant across multiple phenological phases (VISIS: OA = 92 %, F1 = 0.93, FPR = 0.13; aRIBIₙᵢᵣ: OA = 70.7 %, F1 = 0.75, FPR = 0.36). Moreover, the proposed two-step approach eliminated false positives and wrong DI estimation effectively in healthy plants. Our findings suggested the potential of feature selection in overcoming the phenological influence in RB detection, as well as the necessity of disease detection before severity quantification in eliminating the pseudo severity estimates in healthy plants.
Why it matches plant phenotyping methodsイネの病害状態（blast severity）をUASハイパースペクトル画像・反射スペクトルから検出および定量する二段階手法を開発・評価しており、フェノタイピング手法が研究の中心である。
abstractMoreover, a two-step method was applied including the RB detection based on a novel feature selection method termed sequential importance selection (SIS) and the DI estimation using linear models based on rice blast indices (RIBIs).
This study introduces a robust framework for generating procedural 3D models of maize (Zea mays) plants from LiDAR point cloud data, offering a scalable alternative to traditional field-based phenotyping. Our framework leverages Non-Uniform Rational B-Spline (NURBS) surfaces to model the leaves of maize plants, combining Particle Swarm Optimization (PSO) for an initial approximation of the surface and a differentiable programming framework for precise refinement of the surface to fit the point cloud data. In the first optimization phase, PSO generates an approximate NURBS surface by optimizing its control points, aligning the surface with the LiDAR data, and providing a reliable starting point for refinement. The second phase uses NURBS-Diff, a differentiable programming framework, to enhance the accuracy of the initial fit by refining the surface geometry and capturing intricate leaf details. Our results demonstrate that, while PSO establishes a robust initial fit, the integration of differentiable NURBS significantly improves the overall quality and fidelity of the reconstructed surface. This hierarchical optimization strategy enables accurate 3D reconstruction of maize leaves across diverse genotypes, facilitating the subsequent extraction of complex traits like phyllotaxy. We demonstrate our approach on diverse genotypes of field-grown maize plants. All our codes are open-source to democratize these phenotyping approaches.
Why it matches plant phenotyping methodsLiDAR点群からトウモロコシ葉の3D形状を再構成し、複雑な植物形質を抽出する手法の開発が中心である。
abstractThis study introduces a robust framework for generating procedural 3D models of maize (Zea mays) plants from LiDAR point cloud data, offering a scalable alternative to traditional field-based phenotyping.
Areca catechu L. (Arecaceae) is an important cash crop in Taiwan(China), Hainan (China), and several South Asian countries. Areca palm yellow leaf disease (YLD) poses a severe threat, leading to reduced yields and eventual plant mortality. The current study differentiates areca palm damage solely based on spectral features. We are the first to integrate LiDAR point clouds with multispectral imagery to distinguish between different damage levels. We standardized the geographic coordinate systems of the LiDAR data and multispectral images, then aligned them using the control point method. During YLD infestation, areca palm leaves turn yellow and eventually fall off. We surveyed over 1000 trees, counting the number of leaves and calculating the proportion of the canopy area covered by yellowing foliage. Based on crown color changes and leaf count, we classified areca palm damage into five levels: healthy, slightly damaged, moderately damaged, severely damaged, and dying/dead. Meanwhile, this study optimized the individual tree segmentation process for areca palm. Using trunk point clouds, we generated seed points and applied region-growing cluster segmentation to achieve a more accurate individual tree profile compared to the traditional watershed algorithm. Based on the segmentation results and crown contours, we extracted the structural and spectral characteristics of individual trees. Multiple algorithms were then applied to classify areca palms into four damage levels: healthy, slightly damaged, moderately damaged, and severely damaged. The classification achieved an overall accuracy of 86.46% and a kappa value of 0.819. The inclusion of LiDAR data improved the overall accuracy by 23.94% compared to using only spectral features. Comparatively, past studies have relied only on spectral differences to determine the area of leaf yellowing and thus further determine the level of damage. In this study, we further noted the structural changes in the canopy caused by leaf abscission, provided a more realistic description of the different damage levels, and constructed more accurate models. The proposed method demonstrates great potential in YLD damage classification and provides an important basis for precise management of plantations.
Why it matches plant phenotyping methodsLiDAR・UAVマルチスペクトル画像から個体樹の構造・スペクトル形質を抽出し、葉黄化と葉落による植物病害の被害レベルを分類する手法が研究の中心であり、個体セグメンテーションの最適化と精度評価も行っている。
abstractWe are the first to integrate LiDAR point clouds with multispectral imagery to distinguish between different damage levels.
This study proposes a real-time fruit volume estimation model based on YOLOv9 (RTFVE-YOLOv9) and binocular stereo vision technology to address the challenges of low automation and insufficient accuracy in fruit volume measurement in complex orchard environments, particularly in scenarios with diverse canopy structures and severe branch-leaf occlusion. The model achieves effective recognition of occluded fruits through the innovative design of a Dual-Scale and Global–Local Sequence (DSGLSeq) module while incorporating a Multi-Head and Multi-Scale Self-Interaction (MHMSI) module to improve the detection performance of small fruit targets. Systematic validation experiments conducted on major economic fruit tree varieties, including apples, pears, pomelos, and kiwifruit, demonstrate that RTFVE-YOLOv9 improved the mean Average Precision (mAP) by 2.1%, 1.6%, 4%, and 3.8% respectively on the four fruit datasets compared to the baseline YOLOv9-c model. The model’s internal working mechanisms were thoroughly revealed through multi-dimensional evaluation, including ablation experiments, Heatmap Analysis, and Effective Receptive Field (ERF) analysis, providing a theoretical foundation for subsequent optimization. The research findings enrich the application theory of computer vision in smart agriculture and provide reliable technical support for achieving precise orchard management.
Why it matches plant phenotyping methods果実の体積という植物器官形質を、YOLOv9と両眼ステレオビジョンで推定する手法を開発・検証しており、画像取得・計算による表現型推定が研究の中心である。
abstractThis study proposes a real-time fruit volume estimation model based on YOLOv9 (RTFVE-YOLOv9) and binocular stereo vision technology
The present critical literature review describes the state-of-the-art innovative proximal (ground-based) solutions for plant disease diagnosis, suitable for promoting more precise and efficient phytosanitary measures. Research and development of new sensors for this purpose are currently a challenge. Present procedures and diagnosis techniques depend on visual characteristics and symptoms to be initiated and applied, compromising an early intervention. Also, these methods were designed to confirm the presence of pathogens, which did not have the required high throughput and speed to support real-time agronomic decisions in field extensions. Proximal sensor-based systems are a reasonable tool for an efficient and economic disease assessment. This work focused on identifying the application of optical and spectroscopic sensors as a tool for disease diagnosis. Biophoton emission, fluorescence spectroscopy, laser-induced breakdown spectroscopy, multi- and hyperspectral spectroscopy (HS), nuclear magnetic resonance spectroscopy, Raman spectroscopy, RGB imaging, thermography, volatile organic compounds assessment, and X-ray fluorescence were described due to their relevant potential. Nevertheless, some techniques revealed a low technology readiness level (TRL). The main conclusions identify HS, single and multi-spatial point observation, as the most applied methods for early plant disease diagnosis studies (88%), combined with distinct feature selection (FeS), dimensionality reduction (DR), and modeling techniques. Vegetation indices (28%) and principal component analysis (19%) were the most popular FeS and DR approaches, highlighting the most relevant wavelengths contributing to disease diagnosis. In modeling, classification was the most applied technique (80%), used mainly for binary and multi-class health status identification. Regression was used in the remaining (21%) scientific works screened. The data was collected primarily in laboratory conditions (62%), and a few works were performed in field conditions (21%). Regarding the study’s etiological agent responsible for causing the disease, fungi (53%) and viruses (23%) were the most analyzed group of pathogens found in the literature. Overall, proximal sensors are suitable for early plant disease diagnosis before and after symptom appearance, presenting classification accuracies mostly superior to 71% and regression coefficients superior to 61%. Nevertheless, additional research regarding the study of specific host-pathogen interactions is necessary.
Why it matches plant phenotyping methods植物病害の早期診断に用いる光学・分光センシング技術を体系的にレビューしており、病害状態という植物表現型の取得・推定手法が中心である。
abstractThe present critical literature review describes the state-of-the-art innovative proximal (ground-based) solutions for plant disease diagnosis, suitable for promoting more precise and efficient phytosanitary measures.
Drought is a major abiotic stress that adversely affects plant growth, physiology, and crop yield. Conventional methods for assessing drought stress tend to be fragmented, targeting either leaves, canopies, or roots, and are often expensive, low-throughput, and lack the ability to provide real-time, whole-plant insights. Addressing these limitations, this study presents a novel, integrated pipeline titled Intelligent Decision Support for Drought Stress (IDSDS) that leverages remote sensing and artificial intelligence (AI) for accurate, real-time monitoring of drought stress across entire plants. The IDSDS pipeline employs low-cost RGB images collected at various growth stages and uses a deep learning-based model to reconstruct hyperspectral data, which is typically costly and complex to obtain. This reconstructed data enables the extraction of key physiological traits, including greenness, saturation, and pigment content. A novel phenotyping metric—Greenness Coefficient (GC), was also proposed, offering precise spatial analysis of drought impact within the plant. The hyperspectral reconstruction model was validated using standard performance metrics such as the correlation coefficient, mean squared error, standard deviation of squared error, and spectral angle mapper (SAM). IDSDS further calculates a comprehensive set of spectral indices (e.g., greenness, leaf pigment, water content) that are closely linked to drought-induced changes. Finally, by integrating these indices with machine learning-based classification models, IDSDS accurately stratifies plant drought stress into seven distinct categories. The results showed that the proposed hyperspectral reconstruction model effectively converts RGB plant images into accurate hyperspectral data, achieving a SAM value between 0.14 and 0.30. This indicates strong spectral similarity, meaning the reconstructed pixel spectra closely align with the reference spectra. The GC, along with other reconstructed spectral indices, supports visual interpretation and enhances the traceability of the system’s outputs, thereby increasing transparency. Additionally, the findings demonstrate statistically significant results (p < 0.001) for these indices in detecting plant drought stress, with a high classification accuracy of 99 % and an average area under the curve (AUC) of 1.00, reflecting precise differentiation of stress across the entire plant. Overall, the study introduces a breakthrough in drought stress monitoring, combining high-throughput and cost-effective RGB imaging with AI to support both scientific research and practical crop management. The IDSDS pipeline lays the groundwork for informed, drought-adaptive decision-making of agricultural crops.
Why it matches plant phenotyping methodsRGB画像から植物のハイパースペクトル情報と生理形質を推定し、乾燥ストレスを分類するIDSDSパイプラインを開発・検証しており、植物表現型取得法が研究の中心である。
abstractthis study presents a novel, integrated pipeline titled Intelligent Decision Support for Drought Stress (IDSDS) that leverages remote sensing and artificial intelligence (AI) for accurate, real-time monitoring of drought stress across entire plants.
Northern corn leaf blight seriously threatens the health of maize crops in Northeast China. The complexity of field environments, coupled with variations in lighting conditions, poses significant challenges for accurate recognition and segmentation of this disease. To address these issues, this study employs CycleGAN networks and other methods to enhance the diversity of the dataset and proposes a lightweight neural network, Yolov5-Mobile-Seg, for the recognition and segmentation of lesion areas caused by Northern corn leaf blight. The Yolov5-Mobile-Seg network uses Mobilev2 as the backbone, integrating the Convolutional Block Attention Module (CBAM) and Fused MobileNet Bottleneck Convolution Module (FusedMBConv). This design enhances the network’s ability to capture critical information from images while minimizing the number of parameters. Additionally, by incorporating the Free Anchors mechanism, the algorithm’s adaptability to varying sizes of lesion areas is enhanced. Experimental results show that this network outperforms other approaches in identifying northern corn leaf blight, achieving an average precision (AP) of 88.8% in the recognition task and 88.0% in the segmentation task. Compared to the original network, the proposed network reduces the number of parameters by 30.6%, while improving the AP of both the recognition and segmentation tasks by 5.1%. This approach facilitates accurate recognition and efficient segmentation of lesion areas, significantly enhancing the precision and speed of damage assessment for northern corn leaf blight in maize fields.
Why it matches plant phenotyping methodsトウモロコシ葉の病斑領域を画像から認識・セグメンテーションし、病害状態を定量化する手法を開発・評価しており、植物フェノタイピング手法が中心である。
abstractproposes a lightweight neural network, Yolov5-Mobile-Seg, for the recognition and segmentation of lesion areas caused by Northern corn leaf blight.
Harvesting is a critical task in the tree fruit industry, demanding extensive manual labor and substantial costs, and exposing workers to potential hazards. Recent advances in automated harvesting offer a promising solution by enabling efficient, cost-effective, and ergonomic fruit picking within tight harvesting windows. However, existing harvesting technologies often indiscriminately harvest all visible and accessible fruits, including those that are unripe or undersized. This study introduces a novel foundation-model-based framework for efficient apple ripeness and size estimation. Specifically, we curated two public RGBD-based Fuji apple image datasets, integrating expanded annotations for ripeness (“Ripe” vs. “Unripe”) based on fruit color and image capture dates. The resulting comprehensive dataset, Fuji-Ripeness-Size Dataset, includes 4,027 images and 16,257 annotated apples with ripeness and size labels. To the best of our knowledge, this is the first published dataset on apples with ripeness and size annotations. Leveraging Grounding-DINO, a foundation-model-based object detector, we achieved robust apple detection and ripeness estimation, with mean Average Precision being 72.8, outperforming other state-of-the-art models in the evaluation on our dataset. Additionally, we developed six size estimation algorithms, made a comprehensive comparison using box-plots, and identified the best algorithm with lowest error and variation. The Fuji-Ripeness-Size Dataset and the apple detection and size estimation algorithms are made publicly available¹1The code and dataset is available at https://github.com/zhukeyi-stan/Fuji_Ripeness_And_Size_Estimation., which provides valuable benchmarks for future studies in automated and selective harvesting.
Why it matches plant phenotyping methodsリンゴの熟度・サイズという植物器官の形質を画像から推定する手法を開発・比較し、データセットとベンチマークも提供しているため、フェノタイピング手法が中心である。
abstractThis study introduces a novel foundation-model-based framework for efficient apple ripeness and size estimation.
Deep learning (DL) models have shown exceptional accuracy in plant disease identification, yet their practical utility for farmers remains limited due to a lack of professional and actionable guidance. To bridge this gap, we developed CDIP-ChatGLM3, an innovative framework that synergizes a state-of-the-art DL-based computer vision model with a fine-tuned large language model (LLM), designed specifically for Crop Disease Identification and Prescription (CDIP). EfficientNet-B2, evaluated among 10 DL models across 48 diseases and 13 crops, achieved top performance with 97.97 % ± 0.16 % accuracy at a 95 % confidence level. Building on this, we fine-tuned the widely used ChatGLM3-6B LLM using Low-Rank Adaptation (LoRA) and Freeze-tuning, optimizing its ability to deliver precise disease management prescriptions. We compared two training strategies—multi-task learning (MTL) and Dual-stage Mixed Fine-Tuning (DMT)—using a different combination of domain-specific and general datasets. Freeze-tuning with DMT led to substantial performance gains, achieving a 33.16 % improvement in BLEU-4 and a 27.04 % increase in the Average ROUGE F-score, surpassing the original model and state-of-the-art competitors such as Qwen-max, Llama-3.1-405B-Instruct, and GPT-4o. The dual-model architecture of CDIP-ChatGLM3 leverages the complementary strengths of computer vision for image-based disease detection and LLMs for contextualized, domain-specific text generation, offering unmatched specialization, interpretability, and scalability. Unlike resource-intensive multimodal models that blend modalities, our dual-model approach maintains efficiency while achieving superior performance in both disease identification and actionable prescription generation.
Why it matches plant phenotyping methods植物病害画像から病害状態を推定するコンピュータビジョン手法の開発・比較検証が中心であり、処方生成も含む実用的なソフトウェア枠組みである。
abstractEfficientNet-B2, evaluated among 10 DL models across 48 diseases and 13 crops, achieved top performance with 97.97 % ± 0.16 % accuracy at a 95 % confidence level.
Hyperspectral imaging (HSI) is a powerful tool for crop phenotypic component analysis, but developing efficient collaborative extraction and modeling methods for image and spectral features is a challenge. This study proposed a novel spatial-spectral fusion detection method for nicotine content utilizing hyperspectral imaging (HSI) and deep learning. Spectra of different regions and multi-channel images extracted by two-dimensional correlation analysis (2D-COS) were employed as inputs. A dual-branch spatial-spectral attention fusion model (DSSAM) was developed to enhance the expression ability of different modal information. Among them, two branches designed based on the residual module were used to extract spatial and spectral features, respectively. For the spectral branch, a multi-region spectral attention encoder (MSAE) was added to dynamically adjust the weights of the spectrum across leaf regions. For the spatial branch, a swin window attention (SWA) module was introduced to improve local feature extraction and spatial structure learning. The results demonstrated that MSAE and SWA could improve the spatial-spectral information fusion ability of the DSSAM. Compared with the dual-branch model without the attention modules, the coefficient of determination (R²) and relative prediction deviation (RPD) of the DSSAM model on the test set increased by 7.85% and 2.64%, respectively, and the Root Mean Square Error (RMSE) decreased by 39.29%. In addition, the DSSAM outperformed traditional chemometric and single-modal models, with a R² of 0.893, a RMSE of 0.289, and a RPD of 3.054. These findings provide a valuable approach for the quality nondestructive detection of cured tobacco leaves and other crop phenotypic components.
Why it matches plant phenotyping methodsハイパースペクトル画像からタバコ葉のニコチン含量を推定する空間・スペクトル融合モデルを開発し、既存モデルとの性能比較で検証しており、表現型取得・推定手法が中心である。
abstractThis study proposed a novel spatial-spectral fusion detection method for nicotine content utilizing hyperspectral imaging (HSI) and deep learning.
Thinning is a critical practice in apple orchard management, directly influencing crop load and fruit quality. To assist automated crop load management, a machine vision system for apple bud detection was developed to be integrated with robotic platforms. The system employed a Kinect Azure sensor for real-time bud detection and branch diameter measurement, utilizing a YOLOv8-based object detection model trained and evaluated across multiple datasets. The evaluation identified the best-performing model by balancing precision, recall, and robustness in the complex and unstructured environments of apple orchards. Several training configurations were assessed, with the selected setup demonstrating a strong balance between precision (68 %), recall (55 %), F1-score (61 %), and mean average precision (mAP: 59 %) across diverse and unstructured orchard environments. This configuration, trained on a combination of FLIR and Kinect Azure data, was chosen for deployment due to its robustness and compatibility with the Kinect Azure sensor in real-world applications. Two proposed imaging methods for branch diameter measurement were validated against manual caliper-based measurements, with statistical analysis revealing no significant differences (p = 0.98). These findings confirm the semi-automated methods as reliable and labor-efficient alternatives for field applications. Additionally, the bud counting algorithm demonstrated accurate tracking and counting of apple buds, effectively avoiding omissions and duplications in real orchard settings. This study underscores the potential of vision systems to revolutionize apple bud thinning, providing a strong foundation for the development of fully automated solutions in precision orchard management.
Why it matches plant phenotyping methodsリンゴ芽の画像検出に加え、枝径という植物形質の画像計測法を開発・手動測定と検証しており、フェノタイピング手法が中心である。
abstracta machine vision system for apple bud detection was developed to be integrated with robotic platforms
The increasing availability of remote sensing (RS) data and advancements in data assimilation (DA) techniques facilitate the non-destructive calibration of mechanistic crop models but necessitate a framework that digitally represents cropping systems and their spectral properties. This study implemented a coupling scheme linking the outputs of a crop model (DSSAT-CROPGRO) with a radiative transfer model (RTMo module in SCOPE). Reflectance data acquired from a multispectral camera mounted on a UAV were assimilated into the coupled model. The DA scheme was tested in an irrigation and fertilization trial with processing tomatoes, a row crop, requiring the adjustment of the model in order to reflect the vegetation and soil pixel proportions. Examining the relative contribution of dynamically updating specific RTMo parameters showed that the coupled model performed better when parameters were adjusted than when using their nominal values. Applying the DA scheme improved the normalized root mean square error (NRMSE) of the Leaf Area Index (LAI) from 59% to 42% and yield from 64% to 35%. The best performance was achieved when the most water-stressed treatment was excluded, resulting in NRMSE of 34% for LAI and 16% for yield. Since the DA scheme presented here performed well at low to moderate water stress, it should be further tested in assimilating space-borne RS data into simulations of large-scale, commercial fields.
Why it matches plant phenotyping methodsUAVマルチスペクトル画像を作物・放射伝達モデルへ同化する手法を実装・評価し、LAIや収量を推定しているため、植物形質取得・推定が研究の中心である。
abstractThis study implemented a coupling scheme linking the outputs of a crop model (DSSAT-CROPGRO) with a radiative transfer model (RTMo module in SCOPE).
The identification of germplasm resources and analysis of phenotypic traits in lettuce hold significant importance for screening superior varieties and advancing genetic research. To uncover phenotypic differences among lettuce types and their dynamic changes during growth, this study utilized high-throughput phenotyping platform (HTPP) to collect and analyze time-series image data of eight lettuce types. Lettuce phenotypic traits were classified into five major categories: morphology, structure, color, texture, and size, which were further subdivided into 24 subcategories. Using multivariate statistical analysis, the study explored the relationships between phenotypic traits and lettuce types. Combining time-series analysis, the study examined the associations among phenotypic traits, lettuce types, and temporal sequences, revealing the dynamic changes of different lettuce types during their growth processes. The results showed that there were significant differences in phenotypic traits among different lettuce types throughout the growth cycle. These differences reflect the genetic characteristics and phenotypic variation patterns among lettuce types, revealing that genotype guides phenotype formation, while different phenotypes directly influence lettuce growth dynamics. Furthermore, by integrating multidimensional phenotypic traits, we constructed a phenotypic fingerprint for lettuce, providing each lettuce plant with a unique identifier that enables rapid detection of phenotypic differences among lettuce individuals and assists in the selection of elite cultivars. This study provides technical support for precise and rapid identification of lettuce germplasm resources, and can be used as the data basis for genetic research.
Why it matches plant phenotyping methodsレタスの高スループット表現型解析プラットフォームを用いた時系列画像の取得・解析が研究の中心で、多次元形質の抽出とフェノタイプ・フィンガープリント構築を扱っているため。
abstractthis study utilized high-throughput phenotyping platform (HTPP) to collect and analyze time-series image data of eight lettuce types.
Alfalfa is the most important forage crop in grassland agriculture. Efficient regional-scale estimates of alfalfa yields are crucial for the precision management of cultivated alfalfa production; however, obtaining rapid and accurate yield assessments remains challenging due to the scarcity of in situ sample data and limited integration of multi-source satellite remote sensing data. To address these issues, this study utilized a yield dataset from 78 sample plots collected during the alfalfa growing season (May–October) and multi-source heterogeneous satellite remote sensing data (Landsat 8, Sentinel-2, and Sentinel-1). By integrating simulated Landsat 8 red-edge bands and applying sample augmentation via the synthetic minority over-sampling technique for regression with Gaussian noise (SMOGN), a high-accuracy framework for estimating cultivated alfalfa yields is proposed based on multi-source remote sensing data and sample augmentation. The study’s key outcomes are as follows. 1) Compared to alfalfa yield estimation models constructed using only Sentinel-2 or Landsat 8 data, incorporating simulated Landsat 8 red-edge bands or Sentinel-1 variables slightly improves the estimation accuracy, with an R² increase of 0.01–0.04, an RMSE decrease of 3.67–8.00 g/m², an RPD increase of 0.02–0.05, and an MAE decrease of 3.34–8.25 g/m². 2) Using the SMOGN algorithm to augment the sample data significantly enhances estimation accuracy, with R² increases of 0.09–0.28 and RMSE decreases of 0.30–38.57 g/m². 3) Integrating spectral bands and vegetation indices derived from the Sentinel-2, Landsat 8, and Sentinel-1 data improves the accuracy of alfalfa yield estimates, with the optimal model constructed using RF algorithm explaining 66 % of yield variation. Overall, multi-source satellite data and sample augmentation techniques represent an extremely promising approach for remote sensing-based alfalfa yield estimation. This study’s findings provide technical support for an innovative method framework for the precision management of cultivated alfalfa production.
Why it matches plant phenotyping methods衛星リモートセンシングと機械学習により、アルファルファの圃場収量を推定する方法枠組みを開発・比較検証しており、収量という植物形質の取得が中心である。
abstracta high-accuracy framework for estimating cultivated alfalfa yields is proposed based on multi-source remote sensing data and sample augmentation
Accurate, real-time, and non-destructive monitoring of fresh tea leaf quality is essential for achieving high standards in tea cultivation. This study aimed to develop robust predictive models for key quality components—tea polyphenols, free amino acids, and the polyphenol-to-amino acid ratio (TP/AA)—by integrating hyperspectral reflectance data and meteorological variables. Hyperspectral data were collected from tea canopy using an ASD HandHeld 2 spectrometer across six representative tea gardens in Jiangsu Province, China, during spring, summer, and autumn. Simultaneously, fresh leaf samples were analyzed for biochemical composition, and corresponding meteorological data were recorded. Sensitive spectral features were extracted using harmonic and wavelet transformations, and feature-based hyperspectral indices were constructed via two- and three-feature combination strategies. Three machine learning algorithms—Random Forest (RF), Least Absolute Shrinkage and Selection Operator (LASSO), and Partial Least Squares Regression (PLSR)—were employed to build predictive models. The results revealed significant seasonal variation in quality components, with tea polyphenols and TP/AA peaking in summer and amino acids in spring. Harmonic and wavelet features outperformed raw reflectance in correlating with quality indicators. Models integrating these features achieved high predictive accuracy for tea polyphenols (R² = 0.59–0.71), free amino acids (R² = 0.65–0.79), and TP/AA (R² = 0.60–0.77) across different periods, which further improved (up to R² = 0.86) using RF and LASSO with the inclusion of meteorological variables. The best model performance was observed in autumn, followed by summer and spring. This research proposes an effective machine learning–based remote sensing approach for non-destructive tea quality assessment, offering practical value for precision management and optimized harvest scheduling in high-quality tea production.
Why it matches plant phenotyping methods茶葉キャノピーのハイパースペクトルから生葉の品質成分を非破壊推定する特徴抽出・機械学習手法を開発し、複数季節・茶園で精度評価しており、表現型取得手法が中心である。
abstractSensitive spectral features were extracted using harmonic and wavelet transformations, and feature-based hyperspectral indices were constructed via two- and three-feature combination strategies.
Plant phenotyping has emerged as a cornerstone for deciphering the complex interplay between plant genetics and environmental factors. To acquire reliable plant phenotypes, it is important to have an accurate, robust, and high-throughput plant phenotyping system platform. During the last decade, the community seems to have a consensus that such systems should be deployed and work in daytime. Many phenotyping results, however, can be inaccurate and unstable in daytime due to rapid changes in lighting and shadows, particularly for vision-based systems. In this work, we build upon a commercial vision-based high-throughput plant phenotyping (HTPP) platform TraitDiscover and customize a nighttime working mode for the platform. In particular, we incorporate several hardware designs tailored to the nighttime environment such as the array-style lighting equipment and the three-axis high-precision automated control system. On the software side, we also integrate state-of-the-art YOLOv8 object detection and K-Net semantic segmentation frameworks to enable high-performance nighttime image analysis. The feasibility and robustness of the system are demonstrated with a case study on rice. To quantify the effectiveness of phenotyping, a high-quality nighttime rice image segmentation dataset is collected, with 360 finely annotated masks of rice plants. Experimental results show that our customized system is able to achieve surprisingly high segmentation performance up to 93.52% mask IoU, which is significantly higher than the metrics reported from daytime phenotyping. From the mage analysis results, we further extract and validate 28 phenotyping parameters related to color, morphology, and texture status. The average R2 between the inferred phenotype parameters and the actual values reached 0.95, demonstrating the reliability and robustness of the system in nighttime phenotyping. Our results and findings may encourage phenotyping practitioners to rethink the current de facto choice of deploying ‘daytime plant phenotyping platforms’.
Why it matches plant phenotyping methods夜間の植物表現型取得プラットフォームをハードウェア・画像解析・データセットとして開発し、イネ形質の抽出精度と頑健性を検証しており、方法が研究の中心である。
abstractwe build upon a commercial vision-based high-throughput plant phenotyping (HTPP) platform TraitDiscover and customize a nighttime working mode for the platform.
Leaf area index (LAI) of chili peppers is an important indicator of plant growth and productivity. Accurate monitoring of LAI is crucial for optimizing growth conditions, and improving crop yields. However, existing remote sensing-based methods for LAI prediction often face low accuracy, especially in high-density vegetation and across different growth stages. These limitations primarily stem from issues such as multicollinearity and the constraints of spectral information processing mechanisms, which affect the stability and accuracy of predictions. To address these challenges, this study proposed a novel prediction method combining a tree-structured parzen estimator (TPE)-optimized two-band vegetation indices (TPE-2BVIs) with random forest regression (RFR). The TPE-2BVIs band optimization algorithm effectively extracts latent spectral information, enhancing LAI prediction stability and accuracy by addressing multicollinearity and nonlinear model complexity. Experimental results show that: (1) hyperspectral images provide more detailed spectral information compared to multispectral images, significantly improving the accuracy of LAI prediction; (2) the use of 2BVIs alleviates the effects of spectral saturation and dynamic variations during different growth stages, further improving prediction accuracy; (3) the proposed TPE-2BVIs band optimization method significantly enhances both the performance and stability of the model. When combined with RFR, the model achieves R² = 0.887, RMSE = 0.520, and NRMSE = 7.554 %. The TPE-2BVIs band optimization algorithm introduced in this study effectively extracts latent spectral information, overcoming the limitations of multicollinearity and the complexity of nonlinear models in traditional methods. This approach significantly improves the stability and accuracy of LAI predictions. The proposed method provides an innovative solution for remote sensing vegetation monitoring and agricultural applications, offering broad potential for estimating phenotypic parameters under diverse environmental conditions.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像から植物形質であるLAIを推定する新規バンド最適化・回帰手法を開発し、精度と安定性を実験的に評価しているため、植物フェノタイピング手法が中心である。
abstractthis study proposed a novel prediction method combining a tree-structured parzen estimator (TPE)-optimized two-band vegetation indices (TPE-2BVIs) with random forest regression (RFR).
The number of spike grains is an important parameter for wheat yield estimation. However, it is challenging to automatically and intelligently count wheat spike grains in the open field environment. In this study, a deep learning framework, called Wheat Spike Grain Point-to-Point Network (WSG-P2PNet), is proposed to count and locate the wheat spike grains in the open field environment. This framework incorporates Efficient Channel Attention (ECA) and Coordinate Attention (CA) after feature extraction and feature concatenation, respectively. These mechanisms effectively highlight the channel features and positional information of the wheat spike grains while suppressing background interference from factors such as stems, leaves and wheat ears. Additionally, standard convolutions in the regression and classification branches are replaced with Spatial and Channel reconstruction Convolutions (SCConv), further enhancing representational capabilities and improving model performance. The results demonstrate that WSG-P2PNet, using VGG19_bn as the backbone network, outperforms five other state-of-the-art methods in terms of accuracy and stability, with an MAE of 1.72 (95% CI 1.67, 1.77), an Acc of 94.93% (95% CI 94.92, 94.93), an RMSE of 2.35 (95% CI 2.26, 2.44), and an R² of 0.8311 (95% CI 0.8218, 0.8404). Ablation experiments illustrate the impact of SCConv, ECA, and CA on the performance of WSG-P2PNet. Notably, WSG-P2PNet still maintains high accuracy in different varieties and growth periods, demonstrating its robustness and generalizability in real-world scenarios. Preliminary experiments also evaluated the correlation between predicted spike grain numbers and wheat yield, with an average Pearson Correlation Coefficient r of 0.7944, indicating a strong positive statistical relationship. The proposed deep learning framework enables rapid and accurate counting and localization of wheat spike grains in the open field environment, which is of great significant for integrated wheat yield estimation.
Why it matches plant phenotyping methodsコムギ穂粒数という植物形態・収量関連形質を、圃場画像から深層学習で計数・位置推定する手法を開発し、比較・アブレーション・異品種および生育期で検証しているため、方法が中心である。
abstracta deep learning framework, called Wheat Spike Grain Point-to-Point Network (WSG-P2PNet), is proposed to count and locate the wheat spike grains in the open field environment.
The number of stems in wheat populations is a fundamental parameter to achieve high yields and a critical agronomic trait in wheat production and variety selection. Although smart agricultural technology can estimate various agronomic parameters, the wheat stem is often obscured by multiple canopy leaves, making estimation challenging. Consequently, the current method to determine the stem number predominantly relies on labor-intensive manual techniques, which are inefficient and significantly influenced by subjective factors. This study proposes the use of augmented reality (AR) glasses as an imaging data acquisition tool to detect the number of wheat stems with high precision based on features from the top canopy and lateral images of wheat clusters. Following a correlation analysis, four color features, Coverage, the texture feature Contrast, and two lateral peak features SI (Peaks1 and Peaks2) of the top canopy image were identified. The study comparatively analyzed the image features from three perspectives for their accuracy in detecting the number of wheat stems. The results indicated a strong correlation between the peak feature (SI) and the number of wheat stems with an R² value above 0.75. The estimation using only canopy image features (CC) resulted in significant errors, where the RMSE was 20 under high-density planting conditions. Using only Peaks1 and Peaks2 yielded higher accuracy in the stem estimation, but uncertainties persisted in some high-density scenarios. Furthermore, the study combined CC and SI for the estimation and used a random forest algorithm to construct a stem estimation model. This model maintained an RMSE below 10, even under high planting densities and below 5 under low densities, which demonstrated high accuracy. This study could provide insights into stem detection for crops similar to wheat and offer a reference for other studies that require hands-free and first-person perspective image acquisition.
Why it matches plant phenotyping methodsARスマートグラスによる多視点画像取得と画像特徴・ランダムフォレストを用いて小麦の茎数を推定する手法が研究の中心であり、植物形質の取得・抽出方法を実質的に開発・評価している。
abstractThis study proposes the use of augmented reality (AR) glasses as an imaging data acquisition tool to detect the number of wheat stems with high precision based on features from the top canopy and lateral images of wheat clusters.
This study proposes a method for maize seedling reconstruction and spatial distribution analysis based on ground-based laser three-dimensional point cloud scanning technology. Using high-precision terrestrial laser scanning (TLS), 3D point cloud data was collected from multiple maize seedling plots, followed by detailed preprocessing and analysis using Trimble Realworks. During the data processing, a regression-based empirical formula, grounded in maize seedling growth characteristics, was proposed. This formula effectively mitigates the challenges of leaf occlusion in densely planted conditions, providing a solution for further point cloud segmentation and analysis. In terms of algorithm design, this study combines DBSCAN and K-means clustering algorithms to effectively overcome the challenges posed by the dense distribution of plants, leaf occlusion, and noise in the point cloud data. Through this multi-clustering approach, plant positions and distributions were accurately identified, row and column spacing calculations were optimized, and a missing plant detection function was implemented. Furthermore, a dynamic plant height calculation method based on ground undulation was proposed, significantly improving the accuracy of plant height measurement and addressing errors caused by terrain variations. Experimental results show that the proposed algorithm achieves high accuracy and robustness across multiple experimental plots, with a plant counting accuracy rate of 98.33%, a row and column spacing deviation rate controlled within 5%, and a plant height calculation accuracy exceeding 97%. These results demonstrate the effectiveness of this method in precise measurement and spatial distribution analysis during the maize seedling stage, providing strong support for precision agriculture. In the future, with further optimization of the technology, this method could be widely applied in agricultural automation and intelligent management.
Why it matches plant phenotyping methodsTLS点云、クラスタリング、遮蔽補正、動的草丈推定を組み合わせたトウモロコシ個体の再構成・形質計測法が研究の中心であり、精度評価も実施している。
abstractThis study proposes a method for maize seedling reconstruction and spatial distribution analysis based on ground-based laser three-dimensional point cloud scanning technology.
Phenotypes, which define an organism’s behaviour and physical attributes, result from the complex interplay of genetics, development, and environment. Predicting future plant traits is mainly challenging due to these dynamic interactions. This work presents AMULET, a modular approach that combines imaging-based high-throughput phenotyping and machine learning to predict morphological and physiological plant traits hours to days before they are visible. Trained with over 30,000 Arabidopsis thaliana plants, AMULET streamlines the phenotyping process by integrating plant detection, prediction, segmentation, and data analysis, enhancing workflow efficiency and reducing time. AMULET achieved impressive performance metrics, including a dice loss of 0.0104 and an IoU score of 0.9948 for the test set, indicating high accuracy in the segmentation task or an R² score of 0.9289 for descriptor estimation. Moreover, Simpler yet Better Video Prediction (SimVP) appeared as the most effective model in predicting plant growth and health status. Using phenotyping images from studies focused on the Arabidopsis thaliana-Pseudomonas syringae pathosystem, AMULET analysed the latent phenom by identifying traits restrictive to human perception and essential to understanding plant response to concrete growth conditions. Techniques like TorchGrad and Gradient-weighted Class Activation Mapping helped to reveal these new hidden traits. AMULET also demonstrated its adaptability by accurately detecting and predicting phenotypes of in vitro potato plants after minimal fine-tuning with just 100 plants. This versatile approach streamlines phenotyping and holds significant promise for improving breeding programs and agricultural management by enabling pre-emptive interventions optimising plant health andproductivity.
Why it matches plant phenotyping methodsAMULETは画像ベースのハイスループット植物表現型取得、セグメンテーション、形質推定、将来予測を統合する手法・ワークフローとして開発・評価されており、方法が研究の中心である。
abstractThis work presents AMULET, a modular approach that combines imaging-based high-throughput phenotyping and machine learning to predict morphological and physiological plant traits hours to days before they are visible.
Green crops like guava, unlike the majority that can be distinctly separated from the background by color contrast, often share color characteristics with the surrounding leaves, making the accuracy of crop detection and further pixel-level prediction decrease in complex natural environment. Therefore, this work took the texture and boundaries of crops as the focus, proposed a Gabor based texture consistency loss and a reverse attention module (RAM). Meanwhile, both a receptive field module (RFM) and a mutual fusion decoder (MFD) were proposed to enhance the utilization of semantic information. Finally, a stepwise prediction refinement method with the deep prediction map as prior information was designed in the model framework, realizing a further enhancement of the inference ability. In the ablation experiments, this work verified the effectiveness of the proposed improvements step by step using Classification Evaluation Metrics and provided the visualization of the reverse attention. In the comparative experiments, this model demonstrated its advantages in contrast to state-of-the-arts such as U-Net, SETR, and SegFormer. The Acc and IoU reached 0.9954 and 0.9420, exceeding those of SegFormer by 0.0087 and 0.0119 respectively, demonstrating its application potential for agricultural robot visual systems. Moreover, to further demonstrate the inference capability of the proposed model, we conducted validation on two open-source building extraction datasets, WHU and MBD, which have similar task difficulties, and achieved significant results.
Why it matches plant phenotyping methodsグアバ果実の画素レベル分割を目的とする画像解析手法を開発・比較検証しており、植物器官の位置・形状を抽出する方法が中心である。
abstractproposed a Gabor based texture consistency loss and a reverse attention module (RAM)
Implementing advanced approaches such as marker-assisted selection into classic breeding programs is critical for increasing genetic gain and meeting the population’s ever-growing food demand. Genome-wide association studies (GWAS) is a well-known method for detecting genetic markers related to various morphological and physiological traits. However, the ability to collect phenotypic data in large panels often limits the feasibility of genetic studies. This study aimed to assess the potential of UAV-borne thermal and hyperspectral imaging for estimating key wheat traits and identifying their genetic architecture. A diversity panel (300 genotypes) was characterized under well-watered and terminal-drought conditions in a rainout shelter facility. Stomatal conductance, leaf area index, and total chlorophyll content were estimated across two growing seasons. A support vector machine model that integrates canopy spectral reflectance and temperature emittance from UAV-borne imagery reduced the root mean square error of stomatal conductance estimation by 28% compared to using canopy reflectance alone. The models were further used to estimate the traits in the entire panel and to detect genomic markers associated with them and their dynamics throughout the season. Altogether, 16 genetic markers associated with alleles conferring these traits were detected, and the most promising markers were validated during an additional growing season. In the validation experiment, both the spectral estimation models and the allelic effect of the markers were consistent with the previous season. This study introduces, for the first time, the use of stomatal conductance estimation based on combining UAV hyperspectral and thermal imagery for genomic mapping. Implementing this integrated approach could promote the development of new climate-resilience wheat varieties to ensure food security worldwide by screening for stomatal conductance, which is not practical with manual measurements.
Why it matches plant phenotyping methodsUAV熱・ハイパースペクトル画像を統合し、気孔コンダクタンス等の植物形質を推定するモデルを開発・検証しており、フェノタイピング手法が研究の中心である。
abstractThis study aimed to assess the potential of UAV-borne thermal and hyperspectral imaging for estimating key wheat traits and identifying their genetic architecture.
This study explores the potential of hyperspectral imaging (HSI) combined with advanced machine learning for early detection of Septoria Leaf Blotch (SLB) in wheat, employing LeafSpec (Wheat Version), a custom-developed handheld hyperspectral scanner optimized for this purpose. Utilizing a temporal-spectral modelling approach with NDVI heatmaps and PCA for disease visualization, the research analyses HSI data from two rounds of experiments, wheat samples across four treatment groups with images collected at different time points from 3 to 19 days after inoculation (DAI) with 1280 images collected in total. The models, developed using Partial Least Squares Regression (PLSR) and Partial Least Squares Discriminant Analysis (PLS-DA), were tested against average spectra from 3 to 17 DAI. Results indicate that the disease can be detected seven days earlier and before visual symptoms appearance estimated by human observation, with the PLS-DA model achieving 96.97 % overall accuracy in temporal classification. Furthermore, images from 12 DAI predict disease progression with an R2 value of approximately 0.7. These findings demonstrate the potential of HSI combined with machine learning to significantly advance early diagnosis and treatment strategies for SLB, suggesting that similar approaches may be beneficial for other crop diseases.
Why it matches plant phenotyping methods小麦葉の病害状態を対象に、カスタム携帯型ハイパースペクトルスキャナーとスペクトル・時間特徴量による早期検出・進展予測手法を開発・評価しており、植物表現型取得が中心である。
abstractemploying LeafSpec (Wheat Version), a custom-developed handheld hyperspectral scanner optimized for this purpose
Biomass estimation of oilseed rape is crucial for optimizing crop productivity and breeding strategies. While UAV-based imaging has advanced high-throughput phenotyping, current methods often rely on orthophoto images, which struggle with overlapping leaves and incomplete structural information in complex field environments. This study integrates 3D Gaussian Splatting (3DGS) with the Segment Anything Model (SAM) for precise 3D reconstruction and biomass estimation of oilseed rape. UAV multi-view oblique images from 36 angles were used to perform 3D reconstruction, with the SAM module enhancing point cloud segmentation. The segmented point clouds were then converted into point cloud volumes, which were fitted to ground-measured biomass using linear regression. The results showed that 3DGS (7 k and 30 k iterations) provided high accuracy, with peak signal-to-noise ratios (PSNR) of 27.43 and 29.53 and training times of 7 and 49 min, respectively. This performance exceeded that of structure from motion (SfM) and mipmap Neural Radiance Fields (Mip-NeRF), demonstrating superior efficiency. The SAM module achieved high segmentation accuracy, with a mean intersection over union (mIoU) of 0.961 and an F1-score of 0.980. Additionally, a comparison of biomass extraction models found the point cloud volume model to be the most accurate, with an determination coefficient (R²) of 0.976, root mean square error (RMSE) of 2.92 g/plant, and mean absolute percentage error (MAPE) of 6.81 %, outperforming both the plot crop volume and individual crop volume models. This study highlights the potential of combining 3DGS with multi-view UAV imaging for improved biomass phenotyping.
Why it matches plant phenotyping methodsUAV多視点画像、3D再構成、SAMによる分割を統合し、アブラナのバイオマス推定手法を開発・比較検証しており、表現型取得と技術性能が研究の中心である。
abstractThis study integrates 3D Gaussian Splatting (3DGS) with the Segment Anything Model (SAM) for precise 3D reconstruction and biomass estimation of oilseed rape.
Dwarf tomatoes, with high edible and ornamental value, require monitoring multiple growth parameters to balance yield and aesthetics. While deep learning has been widely applied in phenotype monitoring, most studies focus on individual growth parameters, overlooking intrinsic relationships. To simultaneously monitor multiple growth parameters across the entire growth stage and different cultivars, this study develops a multi-modal multi-task phenotype monitoring network for dwarf tomatoes (TomPhenoNet). The network model utilizes top-view RGB-D images to evaluate four key growth parameters: height, leaf area, fresh weight, and the number of red fruits. TomPhenoNet generates mask images, fruit detection features, and the number of detected fruits based on RGB images. By fusing RGB-D images, mask images, and fruit detection features, and introducing the cross-stitch network, the network predicts plant height, leaf area, and fresh weight. The predicted values are further used to generate the dynamic occlusion coefficient, adjusting the number of detected fruits to accurately predict the number of red fruits. Results reveal that TomPhenoNet achieves high prediction performances, with R² values of 0.828, 0.930, 0.945, and 0.881 for plant height, leaf area, fresh weight, and the number of red fruits, respectively. Ablation experiments show that the cross-stitch network and fruit detection features improve the prediction performances of growth parameters, with TomPhenoNet combining both modules performing best. Feature importance analysis indicates the network model captures plant growth characteristics and corrects the impact of leaf occlusion from the top view. This study promotes accurate tomato monitoring and provides data support for optimizing cultivation strategies.
Why it matches plant phenotyping methodsトマトの複数形質をRGB-D画像から推定するマルチモーダル・マルチタスク手法を開発し、性能評価とアブレーション実験まで行っており、表現型取得・推定法が研究の中心である。
abstractthis study develops a multi-modal multi-task phenotype monitoring network for dwarf tomatoes (TomPhenoNet).
Vitreousness serves as a crucial visual indicator of grain hardness and is of paramount importance in the wheat industry due to its substantial influence on both milling and baking quality. Consequently, it is regarded as a fundamental criterion for assessing wheat quality and determining its market value. This study evaluates the efficacy of hyperspectral imaging (HSI) in classifying the grains of thirty-six wheat varieties as either vitreous or non-vitreous, focusing on classification performance across different spectral regions, including Visible (Vis), Visible-Near Infrared (Vis-NIR), and Short-Wave Infrared (SWIR). To achieve this, Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and Artificial Neural Networks (ANN) were utilised to classify grains according to their vitreousness. The results revealed that vitreous kernels were more readily classified than non-vitreous kernels, with classification accuracies of 93.01 % and 83.13 %, respectively. The highest F1 score for the test set, 85.26 %, was attained in the Vis-NIR range by SVM. Region of interest (ROI) selection improved non-vitreous classification by up to 3 %, particularly in the Vis and Vis-NIR regions. Furthermore, five critical wavelengths (540, 636, 476, 588, and 489 nm) in the Vis range were identified using the Minimum Redundancy Maximum Relevance (mRMR) approach. Notably, the reduced set of wavelengths yielded classification accuracies comparable to those obtained using the full spectrum, achieving an accuracy of 93.56 % for vitreous grains and 77.90 % for non-vitreous grains. These findings highlight the potential of HSI, particularly within the Vis region, for the non-destructive classification of wheat grain vitreousness, with colour information emerging as a vital factor in the classification process.
Why it matches plant phenotyping methods小麦粒の硬質性に関わる可視性を、ハイパースペクトル画像と機械学習で非破壊分類する手法を評価しており、植物器官の状態推定が研究の中心です。
abstractThis study evaluates the efficacy of hyperspectral imaging (HSI) in classifying the grains of thirty-six wheat varieties as either vitreous or non-vitreous
Above-ground biomass (AGB) is a key indicator for evaluating maize growth dynamics and yield. Although the remote sensing methods have demonstrated utility in biomass estimation, they often overlook the fundamental heterogeneity in spectral contributions between photosynthetic (primarily leaves) and non-photosynthetic organs (stems, ears, and tassels). In this study, we present a methodology to predict AGB by integrating spectral remote sensing and allometric growth theory. We first demonstrate that leaf organs predominate in determining canopy spectral characteristics, with non-leaf components exhibiting minimal influence on spectral signatures. Building on this theoretical foundation, we developed a two-stage estimation framework that first quantifies leaf biomass using canopy spectral indices and subsequently predicts non-leaf organ biomass through stage-specific allometric growth relationships. Results demonstrate the substantial improvements in estimation accuracy, with the framework achieving an R² of 0.79 and RMSE of 300.09 g/m². Compared to direct spectral estimation of total AGB, we significantly improve prediction accuracy, demonstrating a 216 % increase in explanatory power and a 46.69 % reduction in error. The framework’s robustness across environmental and temporal scales validates its theoretical foundation and practical utility. These findings advance our understanding of biomass allocation dynamics while providing a rigorous approach for non-destructive biomass estimation in maize cultivation systems.
Why it matches plant phenotyping methodsトウモロコシの地上部バイオマスという植物形質を、スペクトルリモートセンシングとアロメトリックモデルで非破壊推定する二段階手法が研究の中心であり、精度改善と頑健性も評価している。
abstractwe developed a two-stage estimation framework that first quantifies leaf biomass using canopy spectral indices and subsequently predicts non-leaf organ biomass through stage-specific allometric growth relationships.
Rapid and accurate monitoring of photosynthetic indicator is of great significance for understanding crop growth and development, and predicting yield. Hyperspectral imagery has become a powerful tool for evaluating photosynthetic capacity due to its non-destructive nature in sensing crop radiation. Most photosynthetic indicators have instantaneous ideal values, which cannot fully reflect the photosynthetic capacity of crop populations in field environments. This study introduces a novel indicator “one day photosynthesis” (ODP) based on the various photosynthetic indicators including net photosynthetic rate (Pn), stomatal conductance (Gs), internal CO₂ concentration (Ci), and transpiration rate (Tr). We performed trend fitting on the time-series photosynthetic indicators obtained at a frequency of two hours, and then calculated the projection area of the fitting curve on the time axis. Later on, the ODP was calculated by assigning weight to the projection area using the CRITIC and correlation method, and the feasibility of ODP was tested using the growth of hundred-grain weight (HGW). Finally, we constructed the ODP estimation model based on canopy hyperspectral data, and further estimated the yield. The results showed that the correlation coefficients between ODP and the growth of HGW were 0.831, 0.882, 0.856, and 0.833 at 10, 20, 30, and 40 days after flowering, respectively. The R² of the ODP estimation model based on hyperspectral vegetation indices (VIs) in the four growth stages were 0.71, 0.83, 0.79, and 0.75, respectively. Moreover, ODP also showed high accuracy and adaptability in different sites, years, sowing dates, and cultivars. We noticed that ODP also has good accuracy in estimating the maize yield, as the R² of estimated yield on the base of measured and estimated ODP was 0.770 and 0.716 respectively. Furthermore, the VIs screened by ODP modeling can also be used for yield estimation, and this VIs screening method is superior to the yield estimation model built based on the correlation between VIs and yield. This study findings provides a novel insight regarding the new ODP indicator that has potential application prospects for efficient estimation of maize photosynthetic capacity and yield.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像から新規指標ODPを推定し、光合成能力と収量を定量化する手法の開発・検証が研究の中心である。
abstractThis study introduces a novel indicator “one day photosynthesis” (ODP)
In order to intelligently and non-destructively estimate soybean yield, the fusion of multicolor space and texture feature parameters were considered to construct a yield prediction model. In this study, the yield prediction model was developed through different stands formed by a management experiment of different planting densities and nitrogen application strategies, and validated through a variety test of 28 soybean varieties. Images of the soybean canopy were collected during the key period for yield (the florescence, podding, and grain-filling stages) by unmanned aerial vehicle (UAV). The multicolor space and texture feature parameters of the RGB images of soybean canopy were extracted for these three periods, and soybean yield prediction models were constructed for the florescence, podding, grain-filling, and multiple growth stages based on different parameters combinations, by using methods of multiple linear regression (SMLR), random forest (RF), and back propagation neural networks (BPNN). The results showed that the color space and texture features of the soybean canopy images exhibited significant differences and different trends during the florescence, podding, and grain-filling stages. Models built with the combinations of texture features (TF) and Hue, Saturation and Value (HSV) parameters had little changes in accuracy, while those incorporating skewed parameters (SP) had better model accuracy. Among all the models, the model combining the SP, TF and HSV parameters demonstrated significantly greater accuracy. The accuracy of models based on individual reproductive stage was lower than those based on entire reproductive period. Thus, the best-performing model was a BPNN model using a combination of the SP, HSV and TF parameters of entire reproductive period as input factors, achieving an R² of 0.765, a prediction accuracy (PA) of the validation set of 88.2 %, and a root mean square error (RMSE) of 430.50 kg/ha. The accuracy of this model in predicting soybean yields across different varieties was PA = 80.4 %, and the RMSE = 514.28 kg/ha. This article provides an effective and low-cost method for accurately estimating soybean yield in the field. This method performs robustly in different varieties and agricultural practices, and has practical value.
Why it matches plant phenotyping methodsUAV画像から色・テクスチャ特徴を抽出し、収量という植物形質を予測する手法を開発・検証しており、表現型取得・推定が研究の中心です。
abstractIn order to intelligently and non-destructively estimate soybean yield, the fusion of multicolor space and texture feature parameters were considered to construct a yield prediction model.
Quantifying the number of Amorphophallus konjac (Konjac) plants can provide valuable insights for yield prediction. Early monitoring of the plant population facilitates timely adjustments in cultivation practices, ultimately leading to improved productivity of Konjac. The majority of research employed deep learning (DL) for plant counting using original images derived from unmanned aerial vehicle (UAV) or ground-based platforms, but this method may lack adaptability to different scenarios and face challenges in achieving plant counting over large areas. This study systematically evaluated the performance of UAV-based original images, the generated orthomosaic, and the combination of both for the detection and counting of the Konjac plant. We proposed an innovative approach by integrating three Convolutional Block Attention Modules (CBAM) into YOLOv5 and utilizing the combined dataset of original images and orthomosaic, which exhibited the highest accuracy performance in Konjac plants recognition (Precision = 94.3 %, Recall = 96.0 %, F1-Score = 95.1 %). Our findings illustrate that the orthomosaic generated from original images acquired via UAV outperformed individual original images in terms of accuracy for counting Konjac plants across expansive areas. This study provides new insight into the recognition and counting of various crop plants across large-scale regions, presenting a practical and efficient approach.
Why it matches plant phenotyping methodsUAV画像とオルソモザイクを用いた植物個体数の検出・計数手法を開発・比較し、YOLOv5改良モデルの性能を評価しているため、植物表現型取得が中心です。
abstractThis study systematically evaluated the performance of UAV-based original images, the generated orthomosaic, and the combination of both for the detection and counting of the Konjac plant.
Technological advances are providing farmers with valuable data about their crops. However, to improve resource use efficiency in agriculture, it is necessary to transform this data into practical information, applicable by farmers and/or technicians in crop management. The objective of this work was to develop a fully automated IoT platform that integrates crop images from RGB cameras with open climate data sources and crop models, to optimize irrigation strategies and enhance crop productivity under varying environmental conditions. To achieve this, the AquaCrop-IoT platform was developed, which integrates the FAO’s AquaCrop model with a custom-build image capture and processing system, used to adjust the green canopy cover (CC) in real-time. Additionally, the platform incorporates weather data from in-situ weather stations, and forecasts and historical weather data from open datasets. Everything is presented in a web application that facilitates its use. The platform has been tested in a wheat crop in southern Spain throughout its growth cycle, demonstrating its potential as a decision support system for irrigation management. Dynamically updating CC values using images captured by the in-situ camera enabled the AquaCrop model to correct potential errors in crop growth estimation by including the effects of adverse factors like pests and diseases that the model cannot simulate. Furthermore, as the developed platform incorporates meteorological data daily, in real-time, it allowed the design of real-time irrigation schedules tailored to the crop in its particular environment and management. This approach improved the estimation of crop water requirements, reducing the amount of recommended irrigation water during the wheat growing season by approximately 32%.
Why it matches plant phenotyping methodsRGB画像から緑色キャノピー被覆率をリアルタイム推定する画像取得・処理系を開発し、AquaCropへ統合した灌漑プラットフォームであり、植物形質取得が中心的な技術貢献です。
abstracta fully automated IoT platform that integrates crop images from RGB cameras with open climate data sources and crop models
The determination of reducing sugars in potatoes is important due to their impact on product quality during industrial processing. The significant variability of these compounds between genotypes presents a challenge to the development of accurate predictive models. This study evaluated the potential of near-infrared hyperspectral imaging (NIR-HSI) for the prediction of reducing sugars in potatoes. For this, a wide range of genotypes (n = 92) from two seasons (2020–2021) was selected. Partial Least Squares Regression (PLSR) and Support Vector Machine Regression (SVMR) methods were used to build the prediction models. Furthermore, interval PLS (iPLS), recursive weighted PLS (rPLS), Genetic Algorithm (GA) and Competitive Adaptive Reweighted Sampling (CARS) were used for relevant wavelength identification to develop less computationally complex models. The best full spectrum model (SNV-PLSR) achieved coefficient of determination and root mean square error values of 0.88 and 0.053 % and 0.86 and 0.057 %, for calibration and external validation, respectively. Variable selection algorithms successfully reduced the dimensionality of the data without compromising the performance of the models. Robust predicted models were built with only 2.65 % (CARS-PLSR) and 3.57 % (iPLS-SVMR) of the total wavelengths. Finally, a pixel-wise prediction was performed on the validation set and chemical images were built to visualise the spatial distribution of reducing sugars. This study demonstrated that NIR-HSI is a feasible technique for predicting reducing sugars in several potato genotypes.
Why it matches plant phenotyping methodsジャガイモ塊茎の還元糖という植物器官形質を、NIR-HSIと回帰モデルで予測・可視化する手法を開発し、外部検証および波長選択による性能評価を行っており、表現型取得が中心である。
abstractThis study evaluated the potential of near-infrared hyperspectral imaging (NIR-HSI) for the prediction of reducing sugars in potatoes.
We present a comprehensive mathematical model to calculate stem water potential in tomato plants cultivated under greenhouse conditions. Stem water potential is one of the variables that determines the growth of fruit as water potential gradients between the fruit and the stem are the driving forces for import of water and solutes into the fruit. Notably, the model integrates growth dynamics, environmental conditions, and plant management strategies to improve the accuracy of water potential estimation throughout the canopy. Environmental factors (i.e., temperature, relative humidity, light irradiance) were implemented at plant compartment levels, allowing for precise microclimate representation. Plant structure was used to calculate water flows and, ultimately, stem water potential by utilizing a hydraulic resistance model. The model was calibrated and validated using data collected from five growing seasons (2020 – 2024). The precision of water potential estimates across different growth stages was improved by including plant morphology dynamics. This, together with discretisation into compartments, allowed for unique realistic predictions for the whole season. Accurate predictions required accounting for growth dependency in root and xylem resistance. Temperature was the main predictor of plant growth for the investigated conditions of tomato production in Belgium. The greenhouse environment and plant management significantly influenced water fluxes and subsequent water potential estimations and should always be considered, especially for whole-season scenarios. Two hypothetical scenarios were analyzed based on 2019 environmental data, exploring the impact of greenhouse management and climate change. Simulations revealed that an increase in the greenhouse minimum temperature set points (+2 °C) had a greater positive effect on yield than a hypothetical climate change scenario with a larger temperature increase (+4 °C). The latter resulted in a higher prevalence of suboptimal growth conditions, presenting a real challenge for efficient future greenhouse management. Additionally, controlling the vapour pressure deficit instead of relative humidity was shown to significantly reduce water demand due to decreased transpiration rates. This water potential model for tomato growth can be used conjointly with fruit growth models for better crop prediction and optimisation of growing conditions. The presented model is modular and extendable, allowing integration not just with fruit growth models, but also potential inclusion of additional plant organs.
Why it matches plant phenotyping methodsトマトの茎水ポテンシャルや形態を推定する数学モデルを開発し、5作期のデータで較正・検証しており、植物状態の取得・推定手法が研究の中心である。
abstractWe present a comprehensive mathematical model to calculate stem water potential in tomato plants cultivated under greenhouse conditions.
The increasing demand for agricultural products and rising production costs have intensified labour shortages in the agricultural sector. Manual harvesting remains essential for products with specific designations, such as wine grapes, where automated solutions cannot match human operators’ dexterity, speed, and care. Minimizing transportation time is also crucial for preserving produce quality and optimizing efficiency. This study aims to optimize harvesting efficiency and vineyard management through the design and implementation of a mobile robotic platform. The platform combines operator dexterity with robotic assistance, continuously tracking operators as they deposit harvested grapes into a harvesting box carried by a robot while gathering data for yield map development. Adaptable to various manual fruit-picking processes, the platform can be integrated into a collaborative harvesting assistance fleet. Field experiments conducted at the Bodegas Terras Gauda (UTM coordinates: 41.95, −8.80, O Rosal, Pontevedra, Spain) vineyard, indicated that operators using robotic assistance reduced their average harvesting time per box by 6 min, increased their total harvested yield by 72.50 kg after two hours (up to 50% more), and reduced manual labour costs by 22.50%. A yield map was developed with high-accuracy GNSS data and an industrial scale mounted on the robot. The map geolocates the weights collected with a maximum variability error of 0.11 kg and successfully expresses grapevine density variability within the same vineyard row. The system preserves produce quality during transportation and significantly eliminates physical strain among operators. These results demonstrate the potential of the robotic platform to improve the efficiency of manual harvesting while maintaining high-quality outcomes.
Why it matches plant phenotyping methodsロボット搭載スケールとGNSSにより収穫重量を地理参照し、ブドウの収量・密度変動を地図化する計測プラットフォームが研究の中心であるため。
abstractA yield map was developed with high-accuracy GNSS data and an industrial scale mounted on the robot.
Canopy spectral information, such as Sun-Induced chlorophyll Fluorescence (SIF) and hyperspectral reflectance, are closely associated with photosynthesis and canopy structure. These spectral indicators provide valuable insights into the actual growth status of crops, thereby guiding management practices in agricultural ecosystems. While considerable efforts have been devoted to simulating the processes of photosynthesis and crop growth, comprehensive and mechanistic modeling of canopy spectral information, integrated with these processes, remains underexplored in traditional crop models. Considering the recent advances in remote sensing observations which are mostly emitted or reflected signals, being able to accurately reproduce the canopy spectra is also advantageous to enhancing the model applicability. In this study, we propose an ecohydrological model (namely the Weishan model) with an integration of a water-carbon-energy fluxes module, a carbon allocation module, a reflectance spectrum module, and a SIF spectrum module for both C₃ (winter wheat) and C₄ crops (summer maize). Comprehensive model calibration and validation have been conducted based on the eddy covariance observations over a typical winter wheat-summer maize rotation cropping cropland in the North China Plain. Validation results highlight the capability and applicability of our ecohydrological model in reproducing the variation of water-carbon fluxes (i.e., evaporation, transpiration, averaged soil moisture, and gross primary productivity), crop growth variables (i.e., leaf area index and end-of-season crop yield), and canopy spectral information (i.e., top-of-canopy SIF, reflectance at near-infrared, red, and blue bands, and vegetation indices). Our model is capable of simulating canopy spectra through mechanistic representations of photosynthesis (e.g., utilizing the Farquhar biochemical model, the Ball-Berry stomatal model, and the energy balance model) and crop dynamics (e.g., phenology, leaf dynamics, carbon allocation and partitioning, biomass accumulation, and yield formation). This comprehensive framework enables the model to effectively disentangle the complex interactions among these processes within a changing environmental context. Furthermore, the model’s ability to accurately reproduce canopy spectra highlights its potential to leverage remote sensing observations to enhance the model performance. We emphasize the functionality and future applicability of our model in advancing ecohydrological and agricultural research.
Why it matches plant phenotyping methods作物のキャノピー分光情報（SIF、反射スペクトル）やLAI・収量などの植物状態を機械論的に推定するモデルを開発し、観測データで較正・検証しており、表現型推定手法が中心である。
abstractwe propose an ecohydrological model (namely the Weishan model) with an integration of a water-carbon-energy fluxes module, a carbon allocation module, a reflectance spectrum module, and a SIF spectrum module
Accurate canopy characterisation is crucial for the targeted application of plant protection products following the variable rate application (VRA) concept. In this study, two different canopy measurement systems were compared: ultrasonic (US) sensors and UAV-based photogrammetry. A specific device was developed to host a series of US sensors that could conduct a fully automatic canopy characterisation of two vine rows in a single pass. The results of canopy characterisation (canopy width, canopy height, leaf wall area, and tree row volume) were compared with those obtained after complete data processing of the images obtained using a multispectral camera embedded on a UAV. Results indicated that no significant differences have been obtained in the definition of main canopy parameters. Field tests indicated that US sensors offered stable canopy height readings but exhibited variability in width measurements due to factors like ground conditions and sensor placement. Compared to with UAV photogrammetry, US sensors provided comparable results for canopy height and width at a lower cost and with less precision. Therefore, the choice between US sensors and UAVs should consider the resolution requirements, cost, and field conditions. Field data were collected from two commercial vineyards in the Penedès region close to Barcelona (Spain). Before this, laboratory tests were performed using an artificial target to achieve an accurate evaluation of the US sensors. Overall, this study highlighted the potential of ground-based sensing systems for precise and repeatable canopy measurements, contributing to improved vineyard management practices and advanced technological integration for agricultural monitoring.
Why it matches plant phenotyping methodsブドウ樹冠の形態形質を取得する超音波センサーとUAV画像法を開発・比較検証しており、フェノタイピング手法が研究の中心である。
abstractA specific device was developed to host a series of US sensors that could conduct a fully automatic canopy characterisation of two vine rows in a single pass.
Tomatoes are an important global crop, and automating the segmentation of leaf diseases is essential for agricultural security. Effective segmentation of these diseases is vital for timely intervention, which can significantly enhance crop yield and reduce pesticide usage. However, challenges such as background interference, tiny diseases, and blurred disease edges pose immense obstacles to the segmentation of tomato leaf diseases. To address these issues effectively, we propose a fusion adversarial segmentation network for tomato disease segmentation named FATDNet. Firstly, to eliminate background interference effectively, we introduce a dual-path fusion adversarial algorithm (DFAA). This algorithm employs parallel dual-path convolution to extract features of leaf disease regions, merging and adversarially processing complex background noise and disease features. Secondly, to enhance the feature extraction of small lesion regions, we employ a multi-dimensional attention mechanism (MDAM). This mechanism allocates weights in both horizontal and vertical directions, subsequently calculating weights in different channels of the feature map. This enhances the dispersion of semantic information through the adoption of diverse weight calculation strategies. Furthermore, to improve the model’s ability to extract features at the edges of leaf diseases, we introduce a Gaussian weighted edge segmentation module (GWESM). This module calculates weight distribution through a Gaussian-weighted function, guiding the model to highlight features at different scales and reduce information loss caused by pooling operations. To demonstrate the superiority of FATDNet, we conduct comparative experiments using a self-built dataset and a public dataset. Experimental results show that FATDNet outperforms nine state-of-the-art segmentation networks. This validates that FATDNet provides a reliable solution for the automated segmentation of tomato leaf diseases.
Why it matches plant phenotyping methodsトマト葉の病斑を画像から自動抽出するセグメンテーション手法を開発し、公開・自作データセットで比較検証しているため、植物病害状態のフェノタイピング手法が中心である。
abstractwe propose a fusion adversarial segmentation network for tomato disease segmentation named FATDNet.
Strawberries have high consumer demand due to their palatability and nutritional benefits. Commercial strawberry production in plant factories with artificial lighting (PFALs) is gaining popularity as a viable strategy for improving economic viability through high-quality fruit production. Accurate information of the optimal harvest date is crucial for optimizing harvesting decisions. While numerous studies utilize deep learning to assess strawberry ripeness, they typically only categorize generalized ripeness levels instead of predicting specific harvest dates, leaving a gap with the practical needs of growers. In this study, we proposed a two-stage multi-feature fusion model for strawberry harvest date prediction and integrated it with a web application to facilitate practical production management in PFALs. The model consists of a fruit segmentation network and a ripeness prediction network. A time-series image dataset of single fruits was constructed to continuously track the ripening process of strawberries, and a five-stage division of strawberry ripeness stages depending on optimal harvest dates was defined. A U-Net based segmentation network with post-processing was developed to automatically extract only the target fruits, which showed a reliable performance with a mIoU of 0.977. A multi-feature fusion network called Triple-Branch Attention Fusion (TBAF) was built to predict the ripeness categories with information on optimal harvest dates. The results showed that the TBAF model with fusion of color, attention-enhanced, and low-level shape features exhibited the highest performance compared to baseline models, with an overall accuracy of 0.859 and an F1 score of 0.859. In addition, a user-friendly web application was developed with the deployment of deep learning models and inspection video processing workflow to support strawberry harvesting in PFALs. Overall, this study demonstrated a prototype approach utilizing deep learning to provide essential information for grower’s decision making in practical strawberry production.
Why it matches plant phenotyping methodsイチゴ果実の画像から成熟段階・最適収穫日を推定するセグメンテーションおよび深層学習ワークフローを開発しており、植物状態の取得・推定手法が中心である。
abstractA U-Net based segmentation network with post-processing was developed to automatically extract only the target fruits
In order to successfully deploy robotic harvesting in open field conditions, the development of an effective machine vision system becomes crucial. In this research, we proposed a novel two-step deep learning model consisting of a modified YOLOv8s and a YOLOv5s-cls to accomplish strawberry detection and pickability classification (whether a mature fruit is pickable by a robot). Firstly, the YOLOv8s was enhanced by incorporating C3x modules and an additional head network structure, specifically tailored for accurate strawberry detection. To further improve training performance, the α-IOU (intersection over union) technique was integrated. Subsequently, the YOLOv5s-cls was utilized to determine suitability of the detected mature strawberries. Through evaluations, Model D (+C3x+head+αIoU), which was a model based on modifying YOLOv8 using the new modules and techniques mentioned above, was found to perform the best among the tested models achieving the highest AP scores of 84.2% in Stage I (immature), 77.8% in Stage II (nearly mature), and 87.8% in Stage III (mature), along with the highest mAP of 83.2%. Overall, this modified model achieved a 2.5% improvement in mAP compared to the same achieved by original YOLOv8s model. Despite a slightly slower inference speed of 8.4 ms per image, Model D maintains real-time capabilities, making it an optimal choice for strawberry detection. Additionally, YOLOv5s-cls was identified as the preferred model for classifying mature strawberries into pickable and unpickable groups, offering a good inference speed of 2.8 ms per image and comparable accuracy with other compared models including YOLOv8s-cls, ResNet 18, EfficientNet-b0, and EfficientNet-b1. Finally, the combined two-step model developed in this study was evaluated in 10 different field scenarios from a completely different strawberry field that was not used in model training and initial testing. In this validation test the machine vision system achieved an AP of 89.0%, 82.0%, and 90.0% in detecting strawberries from Stage I, II, and III while the classification accuracy was 100.0% in unpickable group and 95.0% in pickable group. The results showed that the developed two-step machine vision system has a potential to improve the overall robotic harvesting system for strawberries grown in open-field conditions.
Why it matches plant phenotyping methodsイチゴ果実の成熟段階と収穫可能性という植物器官の状態を推定する機械視覚法を開発し、別圃場で検証しており、単なる収穫対象の位置検出を超えた中心的なフェノタイピング手法である。
abstractwe proposed a novel two-step deep learning model consisting of a modified YOLOv8s and a YOLOv5s-cls to accomplish strawberry detection and pickability classification (whether a mature fruit is pickable by a robot).
Agriculture is the foundation of life that faces numerous daily attacks from nature and living organisms. A major challenge for farmers is timely plant disease identification, which is crucial to prevent productivity losses and the production of poor-quality products. Researchers have recently been focusing on automating the plant leaf disease recognition process using computer vision and machine learning techniques. More importantly, the recent developments in deep learning have significantly advanced the field of plant leaf disease recognition. Regardless of these advancements, significant challenges remain in automatic leaf disease recognition, and researchers are continuing to seek better performance, in-field applicability, and compatibility with resource-constrained devices. This survey provides a comprehensive overview of real-world and laboratory datasets, feature extraction methods, deep learning frameworks, limitations, recommendations, and future directions for deep plant leaf disease recognition. It offers a detailed comparative analysis of various deep learning models applied to different datasets, preprocessing techniques, and data collection methods. This work also highlights the need for an ideal dataset and explores future directions like the Internet of Things integration, Explainable AI, and Smart Farming, which previous surveys have not covered. The primary aim of this survey is to assist researchers in understanding state-of-the-art plant leaf disease recognition techniques, support farmers in the field of plant pathology, address limitations, provide recommendations and outline future directions.
Why it matches plant phenotyping methods植物葉の病徴・病害状態を画像と機械学習で認識する手法を対象とした包括的サーベイであり、データセット、特徴抽出、モデル比較、前処理、データ収集を中心に扱うため、植物フェノタイピング手法レビューに該当します。
abstractThis survey provides a comprehensive overview of real-world and laboratory datasets, feature extraction methods, deep learning frameworks, limitations, recommendations, and future directions for deep plant leaf disease recognition.
Globally, potatoes are the fourth most produced food crop, and in the United Kingdom alone, they generated approximately £705 million in 2022. However, to achieve the United Nations (UN) Sustainable Development Goals (SDG), potato farmers need to sustainably increase yields to address the growing demand for both food and land. Crop yield can be affected by various factors, including disease, pests, and nutrient deficiencies. To tackle these challenges and optimise yields, researchers have leveraged remote sensing platforms for high-throughput non-destructive phenotyping. Data collected from these platforms can be used to develop machine learning (ML) models aimed at addressing the aforementioned issues. To summarise recent developments in ML models applied to potato plant phenotyping, a systematic review of journal articles from the last seven years was conducted. This review underscored the advantages of Deep Learning (DL) approaches and the rising trend of Convolutional Neural Network (CNN)-based architectures, while also noting the limited availability of data for training these models. This review is intended to benefit researchers and farmers by providing an up-to-date review of ML models in potato plant phenotyping.
Why it matches plant phenotyping methodsジャガイモ植物フェノタイピングに用いるリモートセンシングと機械学習モデルを体系的にレビューしており、フェノタイピング手法のレビューが中心です。
titlePotato plant phenotyping and characterisation utilising machine learning techniques: A state-of-the-art review and current trends
Currently, rice transplanters are extensively employed for the mechanized cultivation of rice seedlings. However, few technologies or systems are available to monitor the operational quality parameters, i.e., the number of missing seedlings, row spacing, plant distance, etc., of rice transplanters. The performance of rice transplanters is directly linked to the growth quality of the seedlings and has a crucial effect on the final yield. Therefore, monitoring the various issues that arise during the operation of rice transplanters in a timely and accurate manner to ensure the quality of the transplanting process is particularly important. To address the above issues, this paper develops a real-time monitoring system for rice transplanters. The system architecture includes embedded devices, an image capture module, and a data upload module. A rice seedling detection model based on an enhanced YOLOv5-Lite neural network is developed, and comparative experimental results demonstrate that the proposed model achieves an mAP@0.5 of 81.9 % for rice seedling detection, which is higher than that of the original YOLOv5-Lite model. We additionally propose a RANSAC-based algorithm to detect rice seeding paths in real time, and the rice seeding path detection results are used to determine the row spacing and plant distance. Specifically, a distance mapping algorithm based on triangular transformations is developed to calculate the row spacing and plant distance in a field. We subsequently calculate the number of missing seedlings between adjacent plants on the basis of the spacing between plants in the same row. Furthermore, a rice seedling tracking and counting algorithm based on an improved ByteTrack algorithm is developed to determine the missing seedling rate, as well as the seeding quantity. We integrate the developed algorithms into a real-time monitoring system and test them at Qixing Farm. The experimental results indicate that the monitoring system achieves an accuracy of 99.2 % for seedling quantity counting and an accuracy of 90.3 % for missing rate counting, with a processing speed of 3.95 frames per second.
Why it matches plant phenotyping methodsイネ苗の検出・追跡・計数、欠株率、条間および株間距離を画像から推定するリアルタイムシステムを開発・評価しており、植物形質取得が研究の中心である。
abstractthis paper develops a real-time monitoring system for rice transplanters
Field / plotMultispectral / hyperspectralLeafPhysiological trait estimation
Precisely quantifying crop nitrogen content is critical for adopting sustainable nutrient management practices. This study offers a comprehensive analysis of using hyperspectral data to accurately measure area-based nitrogen content (N) in almond trees at the leaf level. We collected spectral data ranging from 400 to 2500 nm of multiple leaves from 190 samples across two orchards spanning two years. Our methodology involves building a hybrid model that merges a physically based model (PROSPECT-PRO) and a data-driven model (multi-output Gaussian process regression), demonstrating exceptional performance in area-based nitrogen prediction, achieving R² values of 0.54 and an RMSE of 0.03 mg/cm² for area-based nitrogen sensing. The hybrid method incorporates synthetic spectra produced through principal component analysis (PCA) and labeled with biochemical traits retrieved by PROSPECT-PRO for training and validation, while the real data was kept unseen for testing. We compared the performance of physically based, hybrid, and data-driven models using R² and NRMSE as metrics. The Partial Least Squares Regression (PLSR) model showed a strong relationship between leaf N and spectral reflectance (R² = 0.75); however, PLSR is prone to bias from the training set and may perform poorly on unseen data. The findings also highlight the importance of the Short-Wave Infrared regionin nitrogen determination, particularly the bands from 2100 to 2200 nm. Additionally, protein content was found to be a more reliable proxy for nitrogen than chlorophyll. By comparing the retrieved leaf traits with ground truth data, we realized that PROSPECT PRO consistently underestimates almond leaf traits such equivalent water thickness (EWT), carbon-based compounds (CBC), and overestimates Nitrogen. Therefore, adjustment factors were determined for these traits that are estimated with PROSPECT-PRO.
Why it matches plant phenotyping methodsアーモンド葉のハイパースペクトル計測と放射伝達・機械学習モデルにより葉窒素含量を推定する手法を開発・比較・検証しており、植物形質取得が研究の中心である。
abstractOur methodology involves building a hybrid model that merges a physically based model (PROSPECT-PRO) and a data-driven model (multi-output Gaussian process regression), demonstrating exceptional performance in area-based nitrogen prediction
By capturing the intricate structural and spectral variations of the plant canopy, we can enhance our ability to model and predict dynamic parameters such as biomass with greater precision. This method not only preserves the plants for continuous monitoring but also provides a scalable and efficient alternative to traditional destructive techniques. The objective of this study was to examine the potential of using image-derived color and geometric plant features to output accurate predictions of three plant biomass accumulation parameters − leaf fresh weight, leaf dry weight, and leaf area for single plant monitoring. Top-view images of a hydroponic ‘Chicarita’ romaine lettuce (Lactuca sativa) crop captured with a color and depth sensor were used as the input of a multiple plants image processing workflow that extracted plant height, canopy morphometric, and color traits at an individual plant level. Two destructive harvest rounds were performed across the plant cycle to measure the observed values for each biomass response given by leaf fresh weight, leaf dry weight and leaf area from two crop cycles. The image-derived traits were used as potential predictors for a simple linear regression used as a baseline model and for two supervised machine learning models (random forest and least absolute shrinkage and selection operator or LASSO regression) to estimate each response. Using extracted depth information, vertical height per plant was estimated with a mean absolute error of 1.51 cm. Random Forest regression models yielded the most accurate predictions on a first harvest round for all three biomass parameters with R² values of 0.74, 0.80, and 0.67 and mean absolute percentage error (MAPE) of 11.77%, 10.16%, and 12.50%. LASSO regression outperformed the other models in a second harvest round with R² values of 0.72, 0.65, and 0.79 and MAPE of 7.79%, 7.76%, and 7.06% for leaf fresh weight, leaf dry weight, and leaf area, respectively. These results suggest that using a selection of canopy descriptors may improve the non-destructive biomass estimation along a lettuce crop cycle, enabling remote monitoring and real-time harvest projections.
Why it matches plant phenotyping methodsRGB-depth画像から植物形質を抽出し、機械学習でレタスのバイオマスを非破壊推定する手法が研究の中心であるため。
abstractimage processing workflow that extracted plant height, canopy morphometric, and color traits at an individual plant level
Crop phenology plays a vital role in field management and yield prediction of crops. The current remote sensing phenology identification methods utilize characteristic points extracted from time-series vegetation index curves to directly correspond to the beginning of growth stages. However, due to the differences between the meaning of remotely sensed phenological dates and ground-observed phenological stages, there may be certain systematic errors in phenology identification using this method. Therefore, the study proposed a novel phenology extraction framework for crop phenological stages, which does not directly correspond to the dates of remote sensing characteristic points extracted from NDVI curves to the ground phenological stages, but establishes functions between them to improve monitoring accuracy, including single-characteristic point translation method (SCTM) and double-characteristic points weighting method (DCWM). The two methods were applied for monitoring the corn phenology in 12 states in the United States using MODIS. The results showed that DCWM had a better performance than SCTM in phenology extraction, and both of them were superior to the conventional method in which the characteristic points directly correspond to the crop phenological stage. Combining the two methods, the optimal RMSEs of Emerged, Silking, Dough, Dented, Mature and Harvested were 5.28 days, 3.44 days, 4.65 days, 3.88 days, 4.09 days and 6.73 days. Compared with the results from direct correspondence method, they were decreased by 80.48 %, 41.69 %, 40.15 %, 22.55 %, 13.53 % and 29.38 %. The R² also increased by 20.51 %, 9.52 %, 8.93 %, 17.74 %, 16.67 %, 3.03 %, respectively. The framework proposed in this study is a further in-depth study based on the extraction of remote sensing characteristic points, which significantly improves the monitoring accuracy of corn phenological stages, and provides technical enlightenment for the precise phenological extraction in future study.
Why it matches plant phenotyping methodsトウモロコシの生育段階という植物状態を、MODIS時系列NDVIから抽出する新規フレームワークを開発し、既存法と比較検証しているため、表現型取得手法が研究の中心です。
abstractTherefore, the study proposed a novel phenology extraction framework for crop phenological stages
In order to characterize plant water deficiencies, this paper presents a custom-developed bioimpedance (BIS) measurement setup designed for in vivo studies that extracts plant leaf parameters using a novel optimization approach based on the particle swarm optimization (PSO) algorithm. The system performs, four-electrode measurements on plant leaves and employs a custom multi-objective cost function to validate parameters for the Double Shell Cole-Cole model. The experiment consisted of two parts: first, pepper plants (Capsicum annuum L.) as a model plant were exposed to drought stress in a light chamber, and their impedance and physiological parameters were measured. In the second part of the experiment, detached pepper leaves were allowed to dry naturally, and impedance measurements were recorded at hourly and tri-hourly intervals. Impedance spectrum measurements from 230 samples (1 Hz to 100 kHz), collected during both experiments, demonstrated that extracellular fluid resistance increases linearly with water loss. The proposed PSO-optimized Double Shell model showed a stronger correlation between extracellular fluid resistance and water loss compared to the widely used Zfit algorithm, which exhibited higher coefficient of variation in the Cole-Cole parameters. Both algorithms showed a significant negative correlation between relative water content and extracellular fluid resistance, but only the proposed PSO-based model detected a relationship between cell membrane capacity and membrane stability index. Additionally, extracellular fluid resistance correlated with photosynthetic efficiency. The results highlight the effectiveness of impedance measurements for assessing plant water status and support the reliability of proposed PSO-based optimization for bioimpedance analysis.
Why it matches plant phenotyping methods植物の水分状態を測定するバイオインピーダンス計測系とPSO最適化モデルを開発・比較検証しており、植物フェノタイプ取得手法が研究の中心である。
abstractthis paper presents a custom-developed bioimpedance (BIS) measurement setup designed for in vivo studies that extracts plant leaf parameters using a novel optimization approach based on the particle swarm optimization (PSO) algorithm.
AppleField / plotFruitPhysiological trait estimation2D/3D reconstructionGrowth / development / phenology
The full view of the apples in the orchard is often obscured by leaves and trunks, making it challenging to accurately determine their ripeness, whilst it is an essential yet difficult task for apple-harvesting robots. Within this context, we propose a novel method to address two critical challenges: ripeness determination and in-field occlusion. The proposed method is trained in a self-supervised manner on a dataset consisting of less than 1% labelled images and the rest of unlabelled images. It is made up of three key parts: a reconstructor, a feature extractor, and a predictor. The reconstructor is designed to reconstruct the missing parts of occluded apples. The feature extractor is introduced to learn ripeness-related features from the vast number of unlabelled images. Unlike the previous approaches classifying the fruit ripeness into several discrete categories, the predictor uses the learned features to generate a continuous ripeness score in the range between 0.0 and 1.0, thus eliminating the need to subjectively pre-define ripeness stages and offering end-users the flexibility to make their own decisions. Experimental results comparing our method to another method with different settings show that our method achieves the best Structural Similarity Index Measure (SSIM) of 0.75 and the second-best Peak-Signal-to-Noise Ratio (PSNR) of 25.36 for reconstructing missing apple parts, whilst using the fewest 86.3M parameters. Besides, our method outperforms 15 other self-supervised methods and even a supervised method in the ripeness score prediction, with the smallest score 0.0127 for fully unripe and the highest score 0.8933 for fully ripe apples. The results demonstrate the potential of our method to be incorporated with in-field robotic systems, enabling them to assess ripeness for selective harvesting effectively. It is helpful to monitor the overall ripeness of large orchards digitally, aid the decision-making processes and advance the goals of smart and precision agriculture.
Why it matches plant phenotyping methodsリンゴの遮蔽画像から連続的な成熟度スコアを推定する自己教師あり画像解析法を開発し、再構成性能と成熟度予測性能を比較検証しているため、植物フェノタイピング手法が中心である。
abstractwe propose a novel method to address two critical challenges: ripeness determination and in-field occlusion
StrawberryField / plotFruitObject detectionGrowth / development / phenology
Strawberry farming requires efficient and adaptable solutions for real-time monitoring to tackle challenges like rapid ripening, perishability, and bad fruit recognition in field applications. However, existing methods often lack the robustness and lightweight design necessary for resource-constrained environments. To address these limitations, we propose MFD-YOLO, a feature-enhanced, distilled neural architecture based on YOLOv7-tiny, for accurate detection of strawberry growth states. First, we developed the MobileNet-MCA (M-MCA) backbone, which enhances feature extraction while significantly reducing redundant computations. Additionally, Partial Convolution (PConv) is incorporated into the E-ELAN module in the neck, improving feature fusion efficiency while reducing parameters. We also proposed the FocusDownNet (FDN) adaptive downsampling method to better capture and fuse multi-scale features. The DepthLiteBlock is designed to replace the CBL module in the prediction layer, further reducing computational complexity. Finally, an adaptive weighted knowledge distillation (AWKD) strategy is employed to balance performance and efficiency. Experimental results demonstrate that MFD-YOLO achieves a mAP@.5 of 97.5%, precision of 96.5%, recall of 93.8%, and an F1 score of 95.0%, operating at 128 FPS with a model size of only 3.58 MB. The proposed model outperforms state-of-the-art models and is successfully deployed on both desktop and Android devices, enabling real-time, efficient detection in resource-constrained environments.
Why it matches plant phenotyping methodsイチゴの生育状態を画像から検出する軽量モデルを開発し、精度・速度・実装性を評価しており、植物状態の取得手法が研究の中心である。
abstractwe propose MFD-YOLO, a feature-enhanced, distilled neural architecture based on YOLOv7-tiny, for accurate detection of strawberry growth states.
The research introduced a novel vegetation index (DCBVI) to effectively monitor dusky cotton bug infestations, addressing the limitations of existing indices, which show low sensitivity to these infestations. Non-imaging hyperspectral data and UAV multispectral data of the cotton canopy were collected during the flowering and boll development stages over two years, and the first-order differential values of each hyperspectral band were calculated. The Pearson correlation coefficient and Relief-F algorithm were used to select the hyperspectral bands associated with insect stress caused by the dusky cotton bug. Four novel vegetation indices were constructed based on difference, ratio, normalization and chlorophyll vegetation index. The vegetation index most strongly correlated with insect pest stress was selected and designated as the Dusky Cotton Bug Vegetation Index (DCBVI). Combined with the existing vegetation indices, GDLR, SVM and CatBoost were used to construct the pest stress classification model of dusky cotton bug, and DCBVI was derived using relevant bands from UAV multispectral remote sensing data. The pest discrimination model for the cotton bug was established, and the pest distribution in cotton fields was visualized. To research the effect of dusky cotton bug infestation on cotton growth, SPAD inversion model for different infestation levels were established by KNN, RF and CatBoost. The Firefly algorithm (FA), Bat algorithm (BA), and Genetic algorithm (GA) were used to optimize both the classification and discrimination models for dusky cotton bug infestation and the cotton SPAD inversion model. The results showed that the correlation between DCBVI and the pest stress level of the dusky cotton bug was 0.833, and the correlation between DCBVI and SPAD was higher than that of existing vegetation indices. The accuracy, precision, recall, and F1 score of the CatBoost dusky cotton bug insect pest classification model, optimized by FA based on DCBVI and the existing vegetation index, reached 93.1%, 93.2%, 92.9%, and 92.8%, respectively. When the BA-CatBoost model was used to invert the SPAD values of the dusky cotton bug under different insect stress, the R² and RMSE of the optimal model were 0.949 and 1.619. This research significantly improves the accuracy of dusky cotton bug pest monitoring and provides a method for constructing a novel sensitivity index based on hyperspectral data for monitoring other insect pests.
Why it matches plant phenotyping methods綿花の害虫ストレスとSPADという植物状態を、ハイパースペクトル・UAVデータから推定する新規植生指数と分類／反演モデルの開発・検証が研究の中心である。
abstractThe research introduced a novel vegetation index (DCBVI) to effectively monitor dusky cotton bug infestations
Rice seedling morphological identification is crucial for assessing the quality of mechanical transplanting, a process that has traditionally relied on subjective and inefficient manual inspections. To address this challenge, this paper introduces a non-contact quality assessment approach for rice mechanical transplanting based on unmanned aerial vehicle (UAV) imagery. Specifically, we propose a semantic segmentation model for rice seedlings based on an improved HRNet architecture, named RSHRNet. Leveraging UAV imagery as the data source, the methodology employs HRNet as the backbone network, facilitating the acquisition of high-resolution feature information through parallel interactions. Then, the object-contextual representations (OCR) module and coordinate attention mechanism are subsequently introduced to comprehensively aggregate contextual feature information and enhance the model’s ability to extract spatial positional information. This approach helps to resolve issues of blurry edges and misclassification related to rice seedling target segmentation. Finally, to validate the effectiveness and advancements of the proposed method, comparative analyses are conducted against classical segmentation algorithms, followed by practical tests. Experimental results demonstrate the outstanding performance of the proposed RSHRNet in the segmentation of rice seedlings captured by UAVs, providing a solid foundation for the evaluation of mechanical transplanting quality.
Why it matches plant phenotyping methodsUAV画像からイネ苗をセグメンテーションし、機械移植品質評価に用いる植物形態の取得手法を開発・比較検証しており、フェノタイピング手法が中心である。
abstractthis paper introduces a non-contact quality assessment approach for rice mechanical transplanting based on unmanned aerial vehicle (UAV) imagery.
With many forests experiencing rapidly declining health, effective management requires increasingly accurate and precise tools to measure tree attributes across scales. Tree health, especially in deciduous species, is strongly correlated with crown condition, specifically crown transparency and dieback. Present-day assessment of these attributes is undertaken using ground-based visual approaches, which can be imprecise and subjective. Here we evaluate the feasibility of applying drone-based digital aerial photogrammetry (DAP) below, within, and above the tree canopy to estimate tree height, diameter at breast height, canopy transparency, and canopy spread. Video imagery was acquired across 18 deciduous trees under leaf-off and leaf-on conditions in Metro Vancouver, British Columbia, Canada, using small, lightweight first-person-view drones. Images were extracted and processed into coloured 3D point clouds using digital Structure-from-Motion Multiview-Stereo photogrammetry. Photogrammetry estimates were compared with field measurements and above-canopy drone-based aerial Light Detection and Ranging (lidar) estimates. The DAP estimates explained significant variance in the field observations and were strongly correlated with both ground-based measurements and lidar estimates, with correlations of height (DAP vs. ground: r = 0.93, RMSE = 1.54 m; DAP vs. lidar: r = 0.94), DBH (DAP vs. ground: r = 0.98, RMSE = 2.90 cm), transparency (DAP vs. ground: r = 0.66, RMSE = 12.61 %), and crown spread (DAP vs. ground: r = 0.88, RMSE = 3.35 m; DAP vs. lidar: r = 0.89). The reconstruction time for each tree using the drone footage was strongly correlated with tree size and seasonal condition, with minimal influence from crown form. This work suggests that first-person view drones can provide accurate information on individual tree attributes associated with tree health, offering a reliable alternative or complement to both ground-based methods and lidar for tree-level measurements in ongoing forest health assessment programs.
Why it matches plant phenotyping methodsドローン画像とSfM-MVSフォトグラメトリにより、樹冠透明度・枯れ込み・樹高・樹冠広がりなどの個体樹木形質を推定し、地上測定およびLiDARと比較検証しているため、手法が中心的である。
abstractHere we evaluate the feasibility of applying drone-based digital aerial photogrammetry (DAP) below, within, and above the tree canopy to estimate tree height, diameter at breast height, canopy transparency, and canopy spread.
Accurate and timely assessment of alfalfa nutritional parameters is crucial for optimizing harvest management, maximizing yield, and ensuring high-quality forage in China’s Hexi Corridor, a key alfalfa-growing region. UAV-based hyperspectral remote sensing offers a nondestructive and efficient method for monitoring these parameters, providing high-resolution data and covering large areas efficiently. Previous studies have faced challenges related to the scarcity and imbalance of hyperspectral samples and the effective selection of spectral bands for evaluating crop nutrients. Additionally, the simultaneous evaluation of multiple nutrient parameters using a common set of spectral bands has rarely been reported. Least Absolute Shrinkage and Selection Operator (LASSO) is an important method for hyperspectral band selection, but its linear fitting process is challenged by the complex relationship between spectral reflectance and plant properties. In this study, we propose a new band selection strategy that identifies the most informative spectral bands and improves model performance by combining the strengths of both LASSO selection of bands and machine learning’s ability to fit complex relationships. To address the issue of imbalanced field samples, we generated high-quality synthetic data using the synthetic minority oversampling technique for regression with Gaussian noise (SMOGN) algorithm. Three machine learning models (ANN, RF, and SVM) were then employed to predict alfalfa nutritional parameters. Our findings show that the proposed synergistic band selection strategy significantly improves model performance, yielding a 14–25 % reduction in RMSE while requiring only 37–59 % of the original spectral bands. By integrating this band selection strategy with the SMOGN method, our optimal model for estimating alfalfa nutrient parameters achieved R² values of 0.92–0.95 and PRMSE values of 5.1–7.1 %. We observed the importance of the spectral regions around 730 nm and 960 nm for predicting alfalfa quality parameters. This finding suggests that existing satellite platforms such as Sentinel-2 and Landsat could improve the accuracy and efficiency of alfalfa quality monitoring by incorporating these specific spectral bands. Overall, our approach provides a robust and transferable framework for improving the accuracy and reliability of remote sensing-based crop quality monitoring, which is important for optimizing the spectral band configurations of future satellite sensors for precision agriculture.
Why it matches plant phenotyping methodsUAVハイパースペクトル画像からアルファルファの栄養形質を推定するため、バンド選択、データ拡張、機械学習モデルを開発・評価しており、形質取得手法が研究の中心である。
abstractwe propose a new band selection strategy that identifies the most informative spectral bands and improves model performance by combining the strengths of both LASSO selection of bands and machine learning’s ability to fit complex relationships.
Leaf shape is of great significance in plant phenotype research. Landmarks method is a widely used morphometric approach, which can comprehensively describe the morphological differences among leaves. However, the selection of landmarks is time-consuming and laborious. An automatic landmarking algorithm is proposed here. Based on conformal mapping, the leaf outline can be transformed into a monotonically increasing function curve, referred to as the ’fingerprint function’. The Dynamic Time Warping (DTW) algorithm was introduced to match landmarks between different leaves. Two leaf datasets were used to validate the algorithm separately in different species and developmental stages. Dataset1 is a public dataset which covers 26 different types of leaves. The average positional difference between automatic and manual landmarks for dataset1 was only 2.95%. Dataset2 consists of cotton leaves collected in the field at various growth stages, and the positional difference for this dataset was all below 5%. These results validate that our algorithm is applicable to a wide range of leaf types and capable of identifying and locating novel features that emerge during leaf growth. The automatic landmarking algorithm can simulate manual landmarking to a great extent. It provides a new approach for automated acquisition of plant leaf shape homology tailored to the research needs of botanists.
Why it matches plant phenotyping methods葉形態のランドマークを自動抽出する手法を開発し、複数の葉データセットで精度検証しているため、植物表現型取得法が研究の中心です。
abstractAn automatic landmarking algorithm is proposed here.
Cotton is an economically important crop cultivated worldwide for textile production. Breeding programs focus on selecting genotypes with favorable traits for high yields. This study introduced 3D Gaussian Splatting (3DGS) to reconstruct high-fidelity three-dimensional (3D) models and developed a segmentation workflow, Cotton3DGaussians, to analyze cotton bolls and extract architectural traits from single plants. Cotton plants were scanned 360° using a smartphone, and photogrammetry was used to estimate camera parameters and reconstruct a sparse point cloud, which was then optimized into a 3DGS model. In Cotton3DGaussians, 2D masks of bolls segmented from four views were mapped to 3D space, and redundant bolls were removed through cross-view clustering. YOLOv11x and a foundation model, segment anything model (SAM), were compared to obtain 2D masks, with YOLOv11x achieving an F1-score 5.9 % higher than SAM. Phenotypic traits such as boll number, volume, plant height, and canopy size were estimated. The 3DGS model exhibited superior rendering quality, achieving a peak signal-to-noise ratio (PSNR) that was 6.91 higher than NeRF. Cotton3DGaussians effectively segmented 3D bolls from multiple views, with mean absolute percentage errors (MAPE) of 9.23 % for boll number, 3.66 % for canopy size, 2.38 % for plant height, and 8.17 % for boll volume compared to LiDAR ground truth. The regression analysis between convex boll volume and boll weight showed a 19.3 % weight error per plant. This study demonstrates the potential of 3DGS for low-cost, high-fidelity 3D modeling, enabling high-resolution phenotyping and advancing cotton breeding programs. The methodology can also be applied to other crops for improved 3D trait measurement research and enhanced productivity.
Why it matches plant phenotyping methods3Dガウシアンスプラッティングと多視点画像・セグメンテーションを組み合わせ、ワタのボールおよび植物体形質を抽出・検証する手法が研究の中心である。
abstractThis study introduced 3D Gaussian Splatting (3DGS) to reconstruct high-fidelity three-dimensional (3D) models and developed a segmentation workflow, Cotton3DGaussians, to analyze cotton bolls and extract architectural traits from single plants.
A decade after the discovery of grapevine red blotch virus (GRBV), there is ample evidence of its detrimental impacts on grapevine physiology, grape composition, and wine production. To mitigate the spread of GRBV in vineyards, roguing is recommended as a disease management response. The imperative to identify and remove diseased vines justifies the development of autonomous scouting. In this study, nearly 700 ground-based hyperspectral images, encompassing both symptomatic and asymptomatic vine canopies, were collected in a Cabernet Franc vineyard during two growing seasons, capturing pre- and post-veraison vine development stages. Spanning 230 bands from visible (VIS) to near-infrared (NIR) domains (510 to 900 nm with 1.7 nm width), canopy spectral signals were isolated from the background through semantic segmentation using U-Net. Simultaneously, the GRBV status of each vine was established in the laboratory through polymerase chain reaction. These two intertwined datasets were used for training various machine learning algorithms and their ensembles. In addition, strategies to reduce dataset size through spectral binning and testing three different feature selection methods (Recursive Feature Elimination, Univariate Feature Selection, and taking into consideration autocorrelation) were explored. Our findings revealed that hyperspectral imagery identified GRBV-infected vines with an accuracy of 75.7 % around harvest, coinciding with the peak of disease symptom expression, utilizing only 19 bands with a 16 nm bin width. Prior to veraison when most vines are asymptomatic, an accuracy of 74.2 % was achieved, employing 5 bands with a 16 nm bin width. This study substantiates the utility of hyperspectral images in the identification of GRBV-infected vines, offering a robust foundation for the development of a streamlined sensing system that holds great promise for the grape and wine industry in effectively scouting vineyards for GRBV.
Why it matches plant phenotyping methodsブドウ樹の感染状態をハイパースペクトル画像と機械学習で推定するセンシング・解析手法を開発・評価しており、植物病害状態の取得が研究の中心です。
Photosynthesis plays a pivotal role in vegetable growth. However, its intricate interplay with plant physiology and environmental factors complicates precise prediction of photosynthetic rates (Pn). Current predictive models primarily focus on environmental influences on photosynthesis, limiting their applicability to leaves exhibiting different physiological traits. To address the challenge, we introduce a novel approach that incorporates chlorophyll fluorescence (ChlF) parameters into a model for predicting Pn across diverse leaf ontogenies. Eggplant leaves were used as experimental samples. We collected 5280 Pn data of leaves with different ChlF parameters under controlled changes in temperature, [CO₂], and light intensity. The Fₒ (initial fluorescence) and Fᵥ/Fₘ (Maximum light energy conversion efficiency of PSII system) were selected as key ChlF indicators using the entropy method. Fₒ and Fᵥ/Fₘ, along with temperature, [CO₂], and light intensity, are key features, while Pn serves as a label, forming a robust modeling dataset. Then, we proposed a Convolutional Neural Network Regression model with Input Encoding and Genetic Algorithm optimization (CNNR-IEGA) to train these environment and fluorescence data and develop the predictive model for eggplant Pn.The results indicate that the model exhibits excellent performance in predicting Pn. On unknown datasets, the root mean square error of the model is only 0.97 μmol·m⁻²·s⁻¹, with a high coefficient of determination reaching 0.99. Compared to models established by other algorithms (including multiple nonlinear regression, support vector regression, and back propagation neural network), the proposed model demonstrates superior performance across training, testing, and validation sets. Furthermore, compared to models without ChlF parameters and those with single ChlF parameters, the proposed model has the highest accuracy. This demonstrates the validity of using fluorescence to characterize crop photosynthetic performance. CNNR-IEGA can serve as a basis for crop growth environment assessment, greenhouse control, and production warning, offering new theories and opportunities for the development of precision agriculture.
Why it matches plant phenotyping methods植物の光合成速度という生理形質を、クロロフィル蛍光・環境データから予測するモデルを開発し、比較検証しているため、表現型取得・推定手法が中心です。
abstractwe introduce a novel approach that incorporates chlorophyll fluorescence (ChlF) parameters into a model for predicting Pn across diverse leaf ontogenies
Uncrewed Aircraft Systems (UAS) are widely used for crop growth monitoring and yield estimation in Precision Agriculture (PA). However, UAS are limited by their relatively small area coverage, high cost, and high data processing needs. High resolution satellites (such as SkySat) are valuable alternatives to UAS in PA. Nonetheless, persistent cloud cover, especially in regions like the South of Texas, limits their utility. This study compared and explored the integration of satellite and UAS imagery for cotton yield estimation. The rationale was to determine the best performing platform among the two, as well as leverage their synergy to mitigate data gaps caused by persistent cloud cover. Using deep learning model, vegetation indices derived from SkySat and P4M (Phantom 4 Multispectral) images were correlated with crop yield data collected during the 2023 season. Results demonstrated that SkySat slightly outperformed P4M in yield estimation, with median accuracies of R² = 0.81 and RMSE = 0.20 ton/ha for SkySat, compared to R² = 0.80 and RMSE = 0.21 ton/ha for P4M. More importantly, when all the SkySat and P4M datasets were combined, accuracy improved by 3 % compared to SkySat-only data. In addition, data collected between 74 and 114 days after planting contributed most significantly to yield prediction. The fusion approach used in this study allows for better spatial and temporal coverage, ultimately enhancing yield prediction reliability in PA. Future research should explore the inclusion of additional sensors such as Synthetic Aperture Radar (SAR) and thermal imagery, which could further improve yield prediction accuracy, especially in cloud-prone regions.
Why it matches plant phenotyping methods衛星・UAS画像と深層学習を用いた綿花収量推定およびプラットフォーム比較が研究の中心であり、植物の収量形質を直接推定するため含める。
abstractThis study compared and explored the integration of satellite and UAS imagery for cotton yield estimation.
Accurate prediction of plant nitrogen content (PNC) in winter wheat is crucial for precise agricultural water and fertilizer management. UAV-mounted sensors provide a non-destructive, real-time method for assessing PNC on a field scale. This study investigates the effectiveness of RGB, multispectral (MS), and hyperspectral (HS) data acquired from UAVs in predicting PNC in winter wheat, along with assessing the model’s robustness across different regions. Spectral data were collected using these sensors during the flowering stage in two different regions. Spectral bands sensitive to PNC were analyzed, and spectral indices were constructed. A Gaussian process regression (GPR) algorithm is employed to integrate spectral indices from different sensors to construct yield prediction models. The performance of the prediction model is analyzed under both equal-weight and unequal-weight integration strategies. Subsequently, the prediction of PNC in winter wheat utilized the dataset from region A as the calibration set, supplemented by samples from region B. The results revealed that integrating data from all three sensors using an unequal weight strategy produced the most optimal predictive performance for both regions. Furthermore, the transfer learning model demonstrated superior performance by incorporating 18 samples from region B into the MS + HS integrated dataset from region A (R² = 0.61, RMSE = 1.30 mg·g⁻¹). This study confirms the potential of unequal weights integration strategy and model updating strategy based transfer learning for PNC prediction across different regions.
Why it matches plant phenotyping methodsUAVのRGB・マルチスペクトル・ハイパースペクトルデータから冬コムギの植物窒素含量を推定するセンシングおよび転移学習手法が研究の中心であり、地域間の予測性能も評価している。
abstractThis study investigates the effectiveness of RGB, multispectral (MS), and hyperspectral (HS) data acquired from UAVs in predicting PNC in winter wheat, along with assessing the model’s robustness across different regions.
This study presents the application of advanced radio frequency (RF) sensors for non-invasive, plant structure-specific water stress monitoring in olive trees (Olea europaea L.), focusing on the cultivars Frantoio and Leccino, known for their differing water-use strategies. The sensing system comprises circular and double-layer rectangular spiral RF sensors, optimised to maximise the quality factor (Q-factor) for enhanced sensitivity. The double-layer design, where one layer is “left-handed” and the other “right-handed,” allows for an increased magnetic field and detection reliability, especially on small branches where signal stability can be challenging. Throughout an 88-day experimental period, olive trees were subjected to full irrigation (FI) and deficit irrigation (DI) treatments. RF sensors were placed on the olive plants trunks and branches to capture plant structure-specific stress responses, with measurements recorded weekly. In the Frantoio cultivar, resonance frequency shifts were pronounced under DI, especially in the trunk and large branches, where notable physiological changes were observed. Correlations were established between resonance frequency data and morpho-physiological indicators such as trunk diameter increment (SDI) and fresh water content (FWC), validating the sensor’s sensitivity to dielectric property variations due to water stress. Anatomical analyses further revealed tissue adaptations in Frantoio under DI, including increased bark and cortex thickness and intensified sclerenchyma fibre formation, indicative of structural changes to support water transport. In contrast, the Leccino cultivar showed minimal frequency variations and lacked significant anatomical alterations, reflecting its conservative water-use strategy and limited sensitivity to stress. This research confirms RF sensors’ potential as precise tools for early water stress detection in olive trees, with an emphasis on sensor placement on main plant structures and sensitivity optimization to enhance accuracy. These findings support the use of RF sensing systems in precision agriculture for sustainable irrigation management, especially in water-limited environments and conditions.
Why it matches plant phenotyping methodsRFセンサーによるオリーブ樹の水ストレス検出法を開発・最適化し、植物構造別の測定と形態・生理指標との相関で妥当性を検証しているため、フェノタイピング手法が中心である。
abstractThis study presents the application of advanced radio frequency (RF) sensors for non-invasive, plant structure-specific water stress monitoring in olive trees (Olea europaea L.)
Remote sensing technology and machine learning methods are being scaled up globally to predict nutrient content based on spectral data. However, there is a lack of rigorous comparison of co-benefit delivery across different factors, which leads to unstable accuracy of the final model owing to insufficient analysis of the factors influencing the prediction model. In particular, for nutrients (e.g. phosphorus), visual symptoms are not obvious or have a certain lag. Therefore, a Three-Level Meta-Analysis model was proposed in this study to extract and analyse a large number of studies, delving into the analysis of various influencing factors and filling the current knowledge gap. Through global synthesis, a Three-Level Meta-Analysis was applied to seven validated datasets of field observations from multispectral remote sensing, including 32 effect sizes, and 46 datasets of field observations from hyperspectral remote sensing, including 630 effect sizes. We thoroughly explored the heterogeneity of a Three-Level Meta-Analysis using the new machine learning method Meta-Forest, while also using Meta-Cart to explore the interaction effects between moderating variables. Through a comprehensive analysis of the literature published over the past 25 years, we determined the importance of matching preprocessing and regression methods for predicting plant phosphorus spectral responses. The combination of pretreatment and regression methods is particularly important for regional-scale phosphorus concentration prediction. Baseline calibration is effective in removing background noise at the regional scale; however, it cannot solve the problem of redundancy between hyperspectral data. It is necessary to combine a regression method that can effectively deal with redundancy between data to improve the accuracy of the model. Nonlinear non-parametric regression can better deal with the complex nonlinear relationship between phosphorus concentration and spectral data and can resist the influence of the quantity and quality of the data itself and the heterogeneity of the study area; therefore, it has excellent prediction ability. The type of spectrometer is crucial for predicting regional phosphorus concentrations using multispectral data, especially when collecting data using drones. This study provides guidance for fully utilising spectral data and establishing a fast, efficient, and non-destructive prediction model for plant P concentrations, revealing the optimal selection of data preprocessing and regression methods.
Why it matches plant phenotyping methods植物リン濃度という明示的な形質をスペクトルから推定する手法について、複数研究・データセットを用いて前処理、回帰法、分光計の性能や組合せを比較・統合したメタ分析であり、フェノタイピング手法の技術評価が中心である。
titlePreprocessing and regression approaches alter the spectral estimation accuracy of plant phosphorus content—A three-level meta-analysis
Rice Bacterial Blight (RBB), caused by Xanthomonas oryzae pv. oryzae (Xoo), is a major rice disease that significantly threatens yield and quality. RBB spreads rapidly under favorable conditions, affects extensive areas, and requires timely, large-scale monitoring due to its narrow window for effective detection. Traditional satellite monitoring methods, which rely on specific remote sensing platforms and extensive ground surveys, often fail to meet the timely and efficient needs of large-scale disease monitoring. To address the limitations of these traditional methods, this study proposes a cross-scale crop disease monitoring approach that integrates unmanned aerial vehicle (UAV) and satellite remote sensing. With RBB disease monitoring in rice as a case study, the inconsistency between different scale remote sensing data is first introduced to align satellite imagery with UAV data. Next, a sensitivity analysis of the original reflectance and disease-related vegetation indices at both scales is conducted to identify features with consistent performance. The minimum redundancy maximum relevance (mRMR) feature selection algorithm is then employed to obtain sensitive feature sets for each scale. Three machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbors (KNN)—were used to develop disease monitoring models at both UAV and satellite scales. The optimal UAV-scale RF model was then applied to the corrected satellite data for cross-scale monitoring. Results indicate that the proposed cross-scale monitoring method achieved an accuracy of 87.78%, a precision of 88.13%, a recall of 87.78%, and an F1-score of 0.88 for the three-class classification of healthy, mildly infected, and severely infected RBB. The method effectively overcomes the reliance on extensive ground survey data typical of traditional large-scale crop disease remote sensing monitoring methods. Furthermore, the developed approach enables the cross-scale transfer of small-scale monitoring models, ensuring timely disease monitoring during outbreaks.
Why it matches plant phenotyping methodsUAV・衛星リモートセンシングと機械学習を統合し、イネの病害状態（健全・軽症・重症）を推定する手法自体が中心的に開発・評価されているため。
abstractthis study proposes a cross-scale crop disease monitoring approach that integrates unmanned aerial vehicle (UAV) and satellite remote sensing
High temperature stress (HT) plays an important role in soybean selection and breeding, it can cause changes in soybean physiological, biochemical and morphological traits, and directly affect the growth and yield of soybean plants. Among these changes, soybean leaves are particularly sensitive to HT during growth and development. It is important to establish a non-destructive method to distinguish the phenotypic differences between soybean plants under HT and control (CK). In this study, data from two years of soybean field trials were used. In the first year, phenotypic information was collected by near-infrared spectroscopy (NIR), microscopic images, and further difference analysis and classification modelling experiments were conducted. In the second year, multispectral image data were collected and analyzed by Soybean high temperature mask autoencoder (SHT_MAE). The SHT_MAE model with a 75% masking ratio achieved an accuracy of 89.16% and an F1-score of 89.18%. Compared with one-dimensional near-infrared and two-dimensional microscopic image fusion models, the classification accuracy of HT and CK is improved by 2.68%. The accuracy of SHT_MAE multispectral model was improved by 16.84% and 6.88%, respectively, compared with models using only NIR or microscopic images. Both spectral and imaging methods effectively distinguish the phenotypic differences between HT and CK soybean leaves, with the multispectral approach based on the SHT_MAE model demonstrating a clear advantage. This study realized the effective distinction of soybean leaves under HT and CK. It provides theoretical support for HT intelligent breeding (using artificial intelligence and data analysis to optimize breeding decisions) and high temperature grade prediction.
Why it matches plant phenotyping methods高温ストレス下のダイズ葉の表現型差を、近赤外分光・顕微鏡画像・マルチスペクトル画像と分類モデルで非破壊的に抽出・比較する手法研究であり、表現型取得と解析が中心である。
abstractIt is important to establish a non-destructive method to distinguish the phenotypic differences between soybean plants under HT and control (CK).
The uneven spatial and temporal distribution of precipitation poses significant challenges to the growth and development of winter wheat. Screening drought-resistant and water-saving winter wheat varieties in water-limited regions is crucial for increasing crop production. However, quickly screening suitable cultivars remains a challenge. Utilizing unmanned aerial vehicles (UAVs) for remote sensing (RS) offers a solution by enabling the prediction of yields, overcoming issues such as the labor-intensive process of manual yield data collection and the difficulty of screening during the growing season. In this study, three types of water treatments were applied to 48 varieties screened in the North China Plain, with each water treatment repeated three times using a randomized block design. The aim is to explore the potential of UAVs for non-destructive yield prediction at various crop growth stages by integrating UAVs-based RS with machine learning, while also screening for drought-resistant and water-saving variety based on predicted yields, actual evapotranspiration (ET) derived from soil water balance and water use efficiency (WUE) at grain yield level. The results indicate that the random forest regression (RFR) model achieved the best prediction results. The optimal data combination of RS, canopy temperature, and data of variety by using RFR yielded the highest coefficient of determination (R²). Additionally, the RFR performs best when using data from the mid-filling stage (single-stage data) and the entire growth stage data (multi-stage data), with R² 0.58 and 0.69, respectively. Among the varieties, Malan 1 and Jimai 765 ranked first and second in both predicted and measured yield assessments, indicating the reliability of the yield prediction model for top-performing varieties. By combining predicted yields from RFR with ET, the screening results demonstrated high consistency between predicted and measured yields. Notably, even yield prediction models with lower R² can still provide satisfactory screening results. These findings will contribute to screening drought-resistant and water-saving winter wheat varieties by UAV. This research accelerates the variety screening process and addresses the conflict between agricultural production and water scarcity in the North China Plain.
Why it matches plant phenotyping methodsUAVリモートセンシングと機械学習による冬コムギの収量予測を中心的に開発・評価し、予測収量を品種スクリーニングに利用しているため、植物フェノタイピング手法に該当する。
abstractThe aim is to explore the potential of UAVs for non-destructive yield prediction at various crop growth stages by integrating UAVs-based RS with machine learning
Fusarium head blight (FHB) is a major wheat disease worldwide, significantly affecting yield and quality. Disease risk assessment and spatiotemporal dynamic prediction are crucial for effective FHB management and control. Although ecological niche models (ENMs) and epidemiological models (EMs) have been widely applied to assess the potential distribution of diseases and simulate their progression, studies integrating these models with satellite remote sensing and meteorological data for crop disease prediction remain limited. To fill this gap, our study developed an integrated prediction framework based on susceptible-exposed-infected (SEI) model. First, remote sensing data extracted host factors, including wheat spatial distribution, key phenological (KPh) stages defined by Day of Year (DOY), and early physiological changes. This information, along with meteorological features, topographic factors, and sampling coordinates, was utilized to construct an ENM based on Maximum Entropy (MaxEnt) algorithm. MaxEnt evaluation results guided input adjustments, ensuring high AUC output to characterize initial infection levels for SEI model. Next, transition rates in SEI model were determined by the coupling of the parameterized response functions of daily temperature, relative humidity, and DOY for KPh stages to mechanize the EM. The mechanistic model (MM), with optimal parameter values derived from sensitivity analysis and optimization, provided a robust prediction of disease occurrence on the sampling day and enabled spatiotemporal dynamic simulation of wheat FHB. The final MM achieved a coefficient of determination of 0.83, mean absolute error of 0.06, root mean square error of 0.072, and classification F1-score of 0.88. The simulated disease progression curve was consistent with the epidemiological characteristics of FHB, exhibiting an S-shaped pattern. These results suggest that integrating remote sensing and meteorological data with MaxEnt and SEI models for FHB prediction holds significant application potential.
Why it matches plant phenotyping methods衛星リモートセンシングでコムギの生理変化・生育段階を抽出し、MaxEntとSEIモデルでFHBの発生状態を予測・検証する枠組みが研究の中心であるため、植物病害状態のフェノタイピング手法として含める。
abstractour study developed an integrated prediction framework based on susceptible-exposed-infected (SEI) model.
In spring and summer, tomato plants grown in greenhouses often experience high levels of (solar) irradiation in a dry atmosphere during the day. On such hot and sunny days, the resulting high transpiration rates greatly deplete the internal water storage pools (i.e., living cells) of the plant, which gives the plant higher daily stress and may result in irreversible plant or fruit damage. To facilitate the replenishment of internal water storage pools of a plant, greenhouse farmers in Belgium and the Netherlands employ a targeted ventilation strategy, which we have dubbed the ‘plant stress-reducing ventilation’ strategy. This is a commonly used, though scientifically largely understudied, technique in greenhouse cultivation. This makes the strategy difficult to master, leaving growers divided on its effectiveness. To better understand and quantify the effects of the stress-reducing ventilation strategy, we equipped tomato plants (Solanum lycopersicum L.) in a commercial Belgian greenhouse with sap flow and stem diameter variation sensors to continuously measure the plant response to the technique. Climate and greenhouse control data were recorded by the climate computer. This plant response was classified and used to generate a decision tree using machine learning, pointing out the most important factors that reduced plant stress when applying the technique. Our approach is novel in the sense that it incorporates plant sensor measurements into a decision tree algorithm for climate control. This integration has proven crucial in comprehending the practical application of the plant stress-reducing ventilation strategy, now better understood from an ecophysiological perspective.
Why it matches plant phenotyping methods植物センサーでストレス応答を連続測定し、機械学習による意思決定木で解析する手法統合が研究の中心であり、単なる生理測定ではない。
abstractwe equipped tomato plants (Solanum lycopersicum L.) in a commercial Belgian greenhouse with sap flow and stem diameter variation sensors to continuously measure the plant response to the technique.
Ensuring accurate predictions of wheat yield and nutritional content is vital for enhancing agricultural productivity and food security. This study aims to improve wheat yield prediction by integrating process-based models (PBM), machine learning (ML), and remote sensing (RS) techniques. Three Decision Support System for Agrotechnology Transfer (DSSAT) wheat models were calibrated and evaluated using field data from three wheat cultivars grown over three seasons in diverse environments. We developed a hybrid PBM-ML-RS approach using polynomial regression to generate iron (Fe) and zinc (Zn) content from nitrogen predictions. The DSSAT wheat models slightly overestimated wheat yield but accurately predicted nitrogen content. The hybrid PBM-ML-RS approach closely estimated Fe and Zn content with a root mean square error (RMSE) of 0.42 t/ha for yield and 0.89 % for nitrogen content. The integration of ML and RS improved the prediction accuracy for Fe and Zn, achieving RMSE values of 0.35 % and 0.28 % respectively. Spatial simulations provided detailed geographic estimations of wheat yield and nutrient content, supporting site-specific management practices. This study demonstrates the potential of combining PBM, ML, and RS for comprehensive yield and nutrition prediction. The findings indicate a modest decrease in protein, Fe, and Zn concentrations with increasing grain yield, exhibiting high variability across different sites and cultivars. Future research should integrate additional data sources to enhance model robustness and applicability to other crops and regions, contributing to sustainable agriculture and food security.
Why it matches plant phenotyping methodsPBM・機械学習・リモートセンシングを統合した小麦の収量および栄養成分推定手法を開発・評価しており、植物形質の取得・推定が中心的です。
abstractWe developed a hybrid PBM-ML-RS approach using polynomial regression to generate iron (Fe) and zinc (Zn) content from nitrogen predictions.
Botrytis cinerea is a fungal pathogen that can affect a wide range of plants, including roses. Resistance against Botrytis is quantitative, making breeding for resistance challenging. To enable proper genetic marker development, high-throughput and objective data on Botrytis sensitivity is essential. Rose petal discs of different cultivars were manually infected with Botrytis and were monitored with hyperspectral imaging using a fully automated spectral imaging setup. Predictive modelling analysis involved both detection of Botrytis and explaining the severity of infection by linking the spectral data to visual scoring by human eye. Furthermore, band selection analysis was performed to detect key spectral bands relevant for Botrytis detection and to facilitate development of lower cost multi spectral systems for detection of Botrytis infected areas in roses. The presented approach can help plant breeders to explore and adapt to new plant phenotyping technologies such as hyperspectral imaging for breeding against biotic and abiotic stresses.
Why it matches plant phenotyping methodsバラ花弁におけるBotrytis感染の重症度をハイパースペクトル画像と予測モデルで検出・推定する方法を開発し、育種向けに評価しており、植物表現型取得が中心である。
abstracthigh-throughput and objective data on Botrytis sensitivity is essential
Field / plotWhole plant / canopy / plot / fieldGrowth / time-series analysisGrowth / development / phenology
Monitoring agricultural land with optical remote sensing offers a valuable tool for estimating crop yield and supporting decision-making for food security. Cropland phenology indicators, such as the start of season (SOS), the end of season (EOS), and the number of growing seasons per year, provide essential information for land managers. While established toolboxes like TIMESAT have been extracting phenological metrics from coarse remote sensing data for two decades, agricultural monitoring applications demand continuous time series of high-resolution data, made possible by the European Union’s Copernicus Sentinel-2 since 2015. Recently, the Copernicus Land Monitoring Service (CLMS) released the pan-European High-Resolution Vegetation Phenology and Productivity (HR-VPP) product suite. We conducted the first comprehensive validation of the analysis-ready SOS and EOS metrics from the VPP dataset of the HR-VPP product over a large set of agricultural fields spanning 10 countries, 14 crop types and 164 growing seasons. Our results demonstrate that the VPP product of the HR-VPP dataset correlates well with the sowing (r2 = 0.75) and harvesting (r2 = 0.56) dates observed in situ. The biases differ between spring (SOS bias: 59 days, EOS bias: 3 days) and winter (SOS bias: 136 days, EOS bias: –44 days) crops, likely due to the suppression of the autumn vegetation signal in the plant phenology index (PPI) by a solar zenith angle-dependent gain factor. We show that other indicators from the HR-VPP Vegetation Indices (VIs) product and re-parameterization of TIMESAT or DATimeS toolboxes are more suitable for winter crop phenology monitoring. This study calls for researchers and practitioners to carefully evaluate the performance of analysis-ready products to ensure their suitability for specific applications, ultimately promoting informed decision-making in agricultural management and food security endeavours.
Why it matches plant phenotyping methods作物フェノロジー指標を抽出するHR-VPP製品を多数圃場・作物で検証し、既存ツールや再パラメータ化手法と比較しているため、植物表現型取得手法が中心である。
abstractWe conducted the first comprehensive validation of the analysis-ready SOS and EOS metrics from the VPP dataset of the HR-VPP product over a large set of agricultural fields spanning 10 countries, 14 crop types and 164 growing seasons.
Accurate assessment of tomato (Solanum lycopersicum) ripeness is essential for the preservation of quality, meeting market demands and ensuring customer satisfaction. However, one of the key problems is accurately assessing the maturity levels of fruit under varying field conditions. Conventional computer vision models such as convolutional neural networks (CNN) demonstrate uneven performance under varying illumination conditions, particularly in arable farms. Further, it requires extensive training that involves fine-tuning entire model parameters and lags in global context learning. To address these issues, this work introduces a novel segmentation framework that integrates the SegFormer architecture with the Low-Rank Adaptation (SegLoRA) module. The proposed model attained significant performance improvement compared to state-of-the-art (SOTA) methods with a mean Intersection over Union (mIoU) of 83.25 %, an F1-score of 90.07 %, a test accuracy of 99.19 %, and a balanced accuracy of 93.88 %. Additionally, the computational cost was reduced by 26.98 % compared to existing SegFormer models. Further, the deployment on an edge computing device confirmed the proposed model’s feasibility in real time, with a minimal prediction delay of 0.065 s per frame. Moreover, its incorporation with an approximate yield estimation algorithm enables precise enumeration of harvestable tomatoes. These results demonstrate the scalability and efficiency of the SegLoRA, adding to the progress in automated ripeness detection and agricultural automation for selective harvesting operations.
Why it matches plant phenotyping methodsトマト果実の成熟度という植物状態を画像セグメンテーションで推定する手法を開発し、精度・計算コスト・リアルタイム実装を評価しているため、植物フェノタイピング手法が中心である。
abstractthis work introduces a novel segmentation framework that integrates the SegFormer architecture with the Low-Rank Adaptation (SegLoRA) module
With the burgeoning global population, the necessity for sustainable and efficient agricultural practices has become paramount. The primary objective of this paper is to develop an accurate and efficient deep learning model for the timely detection of plant diseases, with a focus on improving crop yield and reducing economic loss. Specifically, this study addresses diseases affecting plant leaves, which present a significant challenge to agricultural productivity. To meet this objective, we introduce LeafConvNeXt, a novel deep learning model tailored to identify plant diseases by meticulously analyzing distinctive features of infected leaves. The secondary objectives include enhancing the interpretability of the model and ensuring its adaptability in resource-constrained environments. LeafConvNeXt integrates convolutional and attention mechanisms, achieving outstanding performance with an accuracy rate exceeding 99% across 52 distinct leaf diseases, outperforming existing contemporary methods. The model’s interpretability is further improved by utilizing LayerCAM, allowing for user-friendly visualization of the diagnostic process. Additionally, its low computational demands and high adaptability make it a practical solution for diverse applications, particularly in its potential integration into intelligent agricultural systems for real-time plant disease monitoring. By emphasizing green energy utilization and regulatory compliance in the era of Artificial Intelligence, LeafConvNeXt lays the groundwork for unmanned farming and a sustainable future in agriculture.
Why it matches plant phenotyping methods葉画像から植物病害を識別する深層学習モデルの開発が研究の中心であり、感染植物の状態を直接推定するため、植物フェノタイピング手法に該当する。
abstractwe introduce LeafConvNeXt, a novel deep learning model tailored to identify plant diseases by meticulously analyzing distinctive features of infected leaves.
Rice blast disease poses a significant threat to rice yield. The disease progresses rapidly once symptoms appear, making timely control challenging. Moreover, once lesions form, the damage becomes irreversible. Existing detection methods often suffer from delays and lack effective strategies for identifying the disease at its asymptomatic stage, hindering early diagnosis. In this study, we collected thermal and optical data from rice canopies at different infection stages and integrated physiological and biochemical analyses to investigate the infection mechanism during the early, asymptomatic phase. Additionally, we employed the SURF feature extraction algorithm to fuse thermal and optical images, developing a preliminary method for identifying asymptomatic rice regions based on thermal signatures. This approach effectively captured the spectral responses of asymptomatic rice and mitigated the limitations of single-sensor detection in early disease identification. By analyzing spectral and temperature characteristics, we applied feature dimensionality reduction techniques to construct early detection models at both the canopy and leaf levels. The models achieved overall classification accuracies (OA) of 92 % and 97 %, respectively, enabling detection 72 h prior to lesion formation. Finally, we designed fixed-point IP and multi-level register-cascade pipeline architecture, implementing low-power FPGA-based edge computing system. The leaf-level detection model deployed on the FPGA achieved an accuracy of 92 %, with a power consumption of 0.076 W and an inference speed of 0.11 ms. This study proposes an effective real-time detection method for identifying early asymptomatic rice blast, thereby facilitating timely disease monitoring and prevention.
Why it matches plant phenotyping methodsイネの無症状病害を熱画像・光学画像から抽出し、検出モデルとFPGA実装まで開発・評価しており、植物状態の取得手法が中心である。
abstractwe employed the SURF feature extraction algorithm to fuse thermal and optical images, developing a preliminary method for identifying asymptomatic rice regions based on thermal signatures.
The tomato is a widely cultivated solanaceous vegetable worldwide and plays a crucial role in meeting human nutritional requirements. Non-invasive, time-dynamic automated representation and analysis of tomato main stems is critical for autonomous monitoring of canopy morphology throughout the entire tomato growth management cycle. Plant growth is influenced by genotype and environment, making naturally curved main stems and mutual shading of the branches and leaves, combined with the limited camera field of view and horizontal camera movement along crop rows, the sensing system observes only discontinuous and curved segments of the main stems. This study proposes an end-to-end YOLOR-Stem approach by optimizing the core components of YOLO v8. First, an innovative method for segmental labelling of main stems using multiple rotating bounding boxes is defined to ensure a precise description. Second, additional angular regression parameters are introduced to capture the orientation and scale information of main stem segments at any angle, overcoming the limitations of horizontal bounding boxes in unstructured field environments. Finally, the Hellinger distance measure is used to quantify the similarity between the predicted and ground truth distributions, integrated into the positive and negative sample matching strategy, loss function computation for rotated bounding boxes, and the prediction box screening during non-maximum suppression. The experimental results demonstrated that YOLOR-Stem (input size of 960 × 960 pixels) with the backbone of EfficientViT-M1 achieved 91.90 % mAP@50, 9.75 M parameters, 35.5GFLOPs, and 10.06 ms inference time. This study enables fast and accurate detection of visible segments of tomato plants, which lays the foundation for intelligent robot-plant interactions such as high-throughput phenotyping, branch and leaf pruning, growth detection, and autonomous harvesting.
Why it matches plant phenotyping methodsトマト主茎の可視セグメントを対象とする回転バウンディングボックス検出手法を開発し、精度・推論時間を評価している。植物形態の取得・解析が中心である。
abstractThis study proposes an end-to-end YOLOR-Stem approach by optimizing the core components of YOLO v8.
The 3D reconstruction of plants is challenging due to their complex shape causing many occlusions. Next-Best-View (NBV) methods address this by iteratively selecting new viewpoints to maximize information gain (IG). Deep-learning-based NBV (DL-NBV) methods demonstrate higher computational efficiency over classic voxel-based NBV approaches but current methods require extensive training using ground-truth plant models, making them impractical for real-world plants. These methods, moreover, rely on offline training with pre-collected data, limiting adaptability in changing agricultural environments. This paper proposes a self-supervised learning-based NBV method (SSL-NBV) that uses a deep neural network to predict the IG for candidate viewpoints. The method allows the robot to gather its own training data during task execution by comparing new 3D sensor data to the earlier gathered data and by employing weakly-supervised learning and experience replay for efficient online learning. Comprehensive evaluations were conducted in simulation and real-world environments using cross-validation. The results showed that SSL-NBV required fewer views for plant reconstruction than non-NBV methods. It achieved IG prediction in 0.0038s, making it over 800 times faster than a voxel-based NBV, and an online learning iteration in 0.099s. SSL-NBV reduced training annotations by over 90% compared to a baseline DL-NBV. Furthermore, SSL-NBV could adapt to novel scenarios through online fine-tuning. Also using real plants, the results showed that the proposed method can learn to effectively plan new viewpoints for 3D plant reconstruction. Most importantly, SSL-NBV automated the entire network training and uses continuous online learning, allowing it to operate in changing agricultural environments.
Why it matches plant phenotyping methods植物の3D再構成に必要な視点選択を自動化する手法を開発し、シミュレーションと実環境で性能評価しているため、植物フェノタイピング手法が中心である。
abstractThis paper proposes a self-supervised learning-based NBV method (SSL-NBV) that uses a deep neural network to predict the IG for candidate viewpoints.
PeachField / plotLiDAR / point cloudLeafMorphology / geometry measurementLeaf traits
Fruit tree canopy leaf area is an important metric for calculating airflow and pesticide dose for accurate variable-rate applications (VRAs) in orchard air-assisted spraying. Existing canopy leaf area calculation models have been established based on the data of a single growth period and the whole canopy leaf area, and it is difficult to meet the precise VRA needs of orchard spraying during whole growth period of fruit trees. In this study, a feature information detection system for fruit tree canopies was designed based on light detection and ranging (LiDAR). Canopy leaf area and LiDAR point cloud detection tests were carried out on peach trees during their whole growth period. The changes of area of individual leaves, leaf number, LiDAR point clouds, and section-based canopy leaf areas and volumes at different growth stages were obtained. Based on least squares regression (LSR) Gaussian fitting and backpropagation (BP) neural network methods, the online calculation models of section-based canopy leaf area was established, and a model modification method was proposed. The R² values of the LSR and BP models increased from 0.865 and 0.863 to 0.906 and 0.898, respectively, and the root-mean-square error (RMSE) decreased from 5110.65 cm² and 5208.74 cm² to 4325.37 cm² and 4600.74 cm², respectively. The accuracy of the model constructed by LSR Gaussian fitting was relatively high, and it was easier to deploy in VRA programs. Compared with those of the existing calculation models, the calculation accuracy and generality of the model constructed in this paper are improved, thus providing model support for the research and development of airflow and pesticide dose on-demand control systems for orchard precision variable-rate spraying.
Why it matches plant phenotyping methodsLiDARを用いた果樹キャノピーの葉面積検出システムと、成長期間全体で葉面積をオンライン推定するモデルを開発・検証しており、植物形質の取得・抽出手法が研究の中心である。
abstractIn this study, a feature information detection system for fruit tree canopies was designed based on light detection and ranging (LiDAR).
Estimating leaf nitrogen (N) status is crucial for site- and time-specific crop N management, and can be accomplished more routinely than ever before with the advent of hyperspectral imaging techniques. Yet, there is still a lack of information about how leaf and canopy N of major crops could be predicted from different regression methods, hyperspectral feature types, and prediction pathways. We conducted field experiments with different N supply for rice, wheat and maize, in China. Features of canopy reflectance (Ref), vegetation indices (VIs), and texture information (Tex) were extracted from acquired hyperspectral images. These features and crop developmental stage (DS) were applied to estimate crop N parameters, using five nonparametric regression algorithms: Partial Least Squares Regression (PLSR), Support Vector Regression (SVR), Random Forest Regression, Deep Neural Network, and Convolutional Neural Network. The performance of PLSR and SVR models was significantly better than that of the others when field samples were limited. Use of feature combination in leaf N prediction was identified necessary from the improved model performance after incorporating the features of Ref, Tex, and DS. The prediction of the mass-based leaf N trait, leaf N concentration, was better than that of the area-based trait, specific leaf N (SLN). Values of SLN and canopy leaf-N content were predicted comparably via themselves direct and indirect methods, although indirect procedures involved more steps requiring the prediction of two or more component traits. These results were discussed in view of making use of available regression-models, features and pathways for best predictabilities so as to improve crop N monitoring for sustainable field N management.
Why it matches plant phenotyping methodsハイパースペクトル画像から作物の葉・群落窒素形質を推定する特徴抽出、回帰モデル比較、予測経路評価が研究の中心であり、実質的な植物フェノタイピング手法の開発・検証に該当する。
abstractFeatures of canopy reflectance (Ref), vegetation indices (VIs), and texture information (Tex) were extracted from acquired hyperspectral images.
Strawberries, as an indeterminate crop, produce fruit multiple times per season, making fruit monitoring and wave-specific yield prediction essential for optimizing harvest planning. This study developed an AI-driven approach to predict next week’s yield using real-time plant image data collected by a machine vision system and environmental data. YOLOv8n was employed to count flowers, immature fruit, and mature fruit per plant, with manual counts used to evaluate the system’s accuracy. The YOLOv8n-based data, combined with weather features, were used to train several AI models for yield prediction. These models included traditional time series machine learning approaches, such as Multiple Linear Regression (MLR) with time lag features, Vector Autoregression (VAR), Gradient Boosting Machines (GBM), Random Forest, and deep learning time-series models, including Long Short-Term Memory (LSTM) and Temporal Convolutional Networks (TCN). Recursive Feature Elimination (RFE) was employed to identify the most relevant features. The performance of these models was evaluated across three strawberry varieties: Sensation, Brilliance, and Medallion. Results showed that MLR outperformed other models for Sensation and Brilliance, with R² values of 0.633 and 0.908, respectively. For Medallion, GBM achieved the best performance with an R² score of 0.848. LSTM, which outperformed TCN, achieved R² scores of 0.522 (Sensation), 0.839 (Brilliance), and 0.740 (Medallion). This AI-driven system automates yield forecasting, reducing labor costs and enabling more efficient harvest planning. The study highlights the potential of combining machine vision and predictive analytics for precise, scalable yield prediction, offering valuable insights for proactive farm management and supply chain optimization.
Why it matches plant phenotyping methods機械視覚で植物ごとの花・未熟果・成熟果を計数し、手動計数で精度評価する手法が収量予測の中心であるため、植物表現型計測・検証を含む。
abstractYOLOv8n was employed to count flowers, immature fruit, and mature fruit per plant, with manual counts used to evaluate the system’s accuracy.
The primary objective of this study was to evaluate the reliability of chili leaf image classification as a method for distinguishing chili germplasm accessions, thereby supporting germplasm conservation efforts. Traditional methods for identifying chili varieties rely on manual leaf observation, which is labor-intensive and error-prone, highlighting the need for advanced fine-grained classification techniques. To address this challenge, we introduce a curated chili leaf database, JNUCLS, to support future research. We evaluate state-of-the-art deep learning methods for chili variety classification and propose a novel approach, LGENetB4CA, which combines a modified EfficientNetB4 model with a LeafGabor filter. The EfficientNetB4 incorporates a Coordinate Attention block following its layers to effectively capture both spatial and channel-wise information. This enables the model to focus on critical regions, such as fine-grained leaf textures, while maintaining a global context. The LeafGabor filter enhances intricate leaf details, such as vein structures, while suppressing noisesuch as blurred shadows, significantly improving input quality. Experiments on the JNUCLS and COLD chili datasets demonstrate high accuracy in chili variety and leaf disease classification, with LGENetB4CA achieving 89.61% accuracy on JNUCLS and 85.90% on COLD chili. These findings highlight the potential of leaf image classification as an effective, cost-efficient tool for exploring phenotypic diversity among chili cultivars. The proposed method also demonstrates promise for broader applications, including plant classification systems, targeted crop management, agricultural product tracking, market analysis, and biodiversity preservation.
Why it matches plant phenotyping methods葉画像から品種差と葉病害を分類する深層学習手法を開発・評価し、データセットも構築しているため、植物表現型取得・分類が中心的です。
abstractwe introduce a curated chili leaf database, JNUCLS, to support future research.
Wheat scab is a fungal disease that threatens wheat yield and quality worldwide. Accurate detection is critical for disease monitoring. However, detection is complicated by the diversity of wheat varieties and growing conditions, as well as challenges such as inconsistent object scales and excessive small objects in unmanned aerial vehicle (UAV) images. Additionally, acquiring large-scale annotated image samples is time-consuming and laborious. To address these challenges, this study proposes a domain adaptive wheat scab detection (DASD) method using UAV. The method first improves the layout images of the diffusion model to generate more realistic labeled target domain images, which are then used to fine-tune the wheat scab detection model. By aligning the feature distributions of the source and target domains, it reduces distribution shift between domains, thereby enhancing the detection model’s adaptation performance without the need for manual annotation of the target domain. Moreover, a Dynamic Multi-feature Fusion Block (DMFB) is designed and embedded into the Real-Time Detection with Transformer (RT-DETR) network to enhance the ability to adapt to varying disease sizes, thereby reducing missed and false detections. Finally, the effectiveness of the proposed method is validated through extensive comparative experiments conducted on multiple constructed datasets. Experimental results show that, without requiring labeled target domain data, the proposed method can improve the Average Precise (AP) value of detection results by 15–23 points compared to the baseline network, providing a feasible solution for large-scale wheat scab detection using UAV. The code is released at https://github.com/yangjie1874/DASD.
Why it matches plant phenotyping methodsUAV画像から小麦赤かび病を検出するドメイン適応手法を開発し、複数データセットで比較検証しており、植物の病害状態の取得・推定が研究の中心です。
abstractthis study proposes a domain adaptive wheat scab detection (DASD) method using UAV
The pear trees, significant both ecologically and economically, play a crucial role in arid and semi-arid areas such as Xinjiang, which make the study of its growth simulation and water use efficiency vital. Few studies have integrated remote sensing with crop growth models to simulate fruit tree growth and plant water transport at the field level. However, plant growth simulations encounter challenges such as uncertain input parameters and regional variability when they are applied to different regions. To address these challenges, we proposed to assimilate inversion data from satellite remote sensing (Sentinel-1, Sentinel-2, and DEM) into the WOFOST model based on Ensemble Kalman Filter (EnKF) techniques to simulate pear tree growth and evaluate water use efficiency at the field scale. We validated this approach at the regional scale by analyzing leaf area index (LAI), yield, and water use efficiency data from 118 pear orchards across five regions during four key growth periods. The results indicated that the NDVI, NDREI, SAVI, and EVI indices were well correlated with LAI. The RIME-LSSVM algorithm enhanced LAI and soil moisture (SM) inversion models, outperforming traditional regression methods. In the four phenological periods, the R² of LAI inversion model ranged from 0.76 to 0.96, NRMSE ranged from 3% to 6.6%, and SM inversion R² values ranged from 0.27 to 0.41, NRMSE values ranged from 10.9% to 16.6%. The results showed that the joint assimilation of SM and LAI into the calibrated WOFOST model significantly improved the simulation performance of regional yield and water use efficiency compared to univariate assimilation and non-assimilation. Specifically, the yield estimation R² increased from 0.37 to 0.58, and the NRMSE decreased from 8.2% to 6.2%. Similar improvements were achieved in the water use efficiency simulations, with R² rising from 0.52 to 0.70 and NRMSE declining from 9.1% to 7.1%. The proposed assimilation method can simulate growth processes and analyze water transport across four critical periods in pear orchards in various regions, aligning with regional observations. The proposed method facilitated the quantitative simulation of pear growth and water transport, providing a promising method for water management in other orchards in arid and semi-arid regions.
Why it matches plant phenotyping methods衛星リモートセンシングによるLAI・土壌水分の反演と、それらをWOFOSTへ同化する手法を開発・検証し、ナシ樹の生育・収量・水利用効率を推定しているため、植物表現型取得・推定が中心的である。
abstractwe proposed to assimilate inversion data from satellite remote sensing (Sentinel-1, Sentinel-2, and DEM) into the WOFOST model based on Ensemble Kalman Filter (EnKF) techniques to simulate pear tree growth and evaluate water use efficiency at the field scale.
Accurately predicting the degree of lodging (breaking or bending of plants due to loss of integrity of the stem or roots) in corn (Zea mays) facilitates variety selection for the seed industry and growers, and provides data to support agricultural insurance claims. Due to various factors such as genetic characteristics and environment, accurately predicting lodging in corn varieties faces challenges in terms of data and models. In this study, we introduce an innovative classification model incorporated high-order relationships with multiple hidden factors to predict the degree of lodging in corn cultivation. Our model integrates a multi-view hypergraph network and incorporates an interpretability analysis component to enhance its predictive capabilities. To effectively capture the complex interactions within corn variety test samples across various dimensions, our model constructs a multi-view hypergraph using meteorological, disease infestation, and phenotype data. This method enables the model to comprehensively identify potential correlations in the test sample data of corn varieties, while considering the multidimensional nature of the problem. Furthermore, to enhance the model’s interpretability, we employ an analytical method to quantify the influence of individual factors on the likelihood and severity of corn lodging events. These insights are then used to fine-tune the model’s predictions. As an empirical evaluation, we applied this model to data collected from 194 corn test sites across mainland China. The results underscore the model’s exceptional performance in predicting the degree of corn lodging, and demonstrate the effectiveness of potential correlation relationships in improving prediction accuracy.
Why it matches plant phenotyping methodsトウモロコシの倒伏という植物状態を予測する解釈可能なマルチビュー・ハイパーグラフ手法が研究の中心であり、表現型データを用いた実データ評価も行っているため、植物フェノタイピング手法として含める。
abstractwe introduce an innovative classification model incorporated high-order relationships with multiple hidden factors to predict the degree of lodging in corn cultivation.
Accurate blueberry fruit phenotyping, including yield, fruit maturity, and cluster compactness, is crucial for optimizing crop breeding and management practices. Recent advances in machine vision and deep learning have shown promising potential to automate phenotyping and replace manual sampling. This paper presented a robotic blueberry phenotyping system, called MARS-Phenobot, that collects data in the field and measures fruit-related phenotypic traits such as fruit number, maturity, and compactness. Our workflow comprised four components: a robotic multi-view imaging system for high-throughput data collection, a vision foundation model (Segment Anything Model, SAM) for mask-free data labeling, a customized BerryNet deep learning model for detecting blueberry clusters and segmenting fruit, as well as a post-processing module for estimating yield, maturity, and cluster compactness. A customized deep learning model, BerryNet, was designed for detecting fruit clusters and segmenting individual berries by integrating low-level pyramid features, rapid partial convolutional blocks, and BiFPN feature fusion. It outperformed other networks and achieved mean average precision (mAP50) of 54.9 % in cluster detection and 85.8 % in fruit segmentation with fewer parameters and fewer computation requirements. We evaluated the phenotypic traits derived from our methods and the ground truth on 26 individual blueberry plants across 17 genotypes. The results demonstrated that both the fruit count and cluster count extracted from images were strongly correlated with the yield. Integrating multi-view fruit counts enhanced yield estimation accuracy, achieving a Mean Absolute Percentage Error (MAPE) of 23.1 % and the highest R² value of 0.73, while maturity level estimations closely aligned with manual calculations, exhibiting a Mean Absolute Error (MAE) of approximately 5 %. Furthermore, two metrics related to fruit compactness were introduced, including cluster compactness and fruit distance, which could be useful for breeders to assess the machine and hand harvestability across genotypes. Finally, we evaluated the proposed robotic blueberry fruit phenotyping pipeline on eleven blueberry genotypes, proving the potential to distinguish the high-yield, early-maturity, and loose-clustering cultivars. Our methodology provides a promising solution for automated in-field blueberry fruit phenotyping, potentially replacing labor-intensive manual sampling. Furthermore, this approach could advance blueberry breeding programs, precision management, and mechanical/robotic harvesting.
Why it matches plant phenotyping methodsロボット多視点撮像、深層学習による果実抽出、収量・成熟度・房密集度の推定を統合した植物表現型計測パイプラインが研究の中心であり、精度検証も実施している。
abstractThis paper presented a robotic blueberry phenotyping system, called MARS-Phenobot, that collects data in the field and measures fruit-related phenotypic traits such as fruit number, maturity, and compactness.
Among various types of forages, Alfalfa (Medicago sativa) is a crucial forage crop that plays a vital role in livestock nutrition and sustainable agriculture. As a result of its ability to adapt to different weather conditions and its high nitrogen fixation capability, this crop produces high-quality forage that contains between 15 and 22 % protein. It is fortunately possible to improve the overall prediction of forage biomass and quality prior to harvest through remote sensing technologies. The recent advent of deep Convolution Neural Networks (deep CNNs) enables researchers to utilize these incredible algorithms. This study aims to build a model to count the number of alfalfa stems from proximal images. To this end, we first utilized a deep CNN encoder-decoder to segment alfalfa and other background objects in a field, such as soil and grass. Subsequently, we employed the alfalfa cover fractions derived from the proximal images to develop and train machine learning regression models for estimating the stem count in the images. This study uses many proximal images taken from significant number of fields in four provinces of Canada over three consecutive years. A combination of real and synthetic images has been utilized to feed the deep neural network encoder-decoder. This study gathered roughly 3447 alfalfa images, 5332 grass images, and 9241 background images for training the encoder-decoder model. With data augmentation, we prepared about 60,000 annotated images of alfalfa fields containing alfalfa, grass, and background utilizing a pre-trained model in less than an hour. Several convolutional neural network encoder-decoder models have also been utilized in this study. Simple U-Net, Attention U-Net (Att U-Net), and ResU-Net with attention gates have been trained to detect alfalfa and differentiate it from other objects. The best Intersections over Union (IoU) for simple U-Net classes were 0.98, 0.93, and 0.80 for background, alfalfa and grass, respectively. Simple U-Net with synthetic data provides a promising result over unseen real images and requires an RGB iPad image for field-specific alfalfa detection. It was also observed that simple U-Net has slightly better accuracy than attention U-Net and attention ResU-Net. Finally, we built regression models between the alfalfa cover fraction in the original images taken by iPad, and the mean alfalfa stems per square foot. Random forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGB) methods have been utilized to estimate the number of stems in the images. RF was the best model for estimating the number of alfalfa stems relative to other machine learning algorithms, with a coefficient of determination (R²) of 0.82, root-mean-square error of 13.00, and mean absolute error of 10.07.
Why it matches plant phenotyping methods近接画像からアルファルファをセグメンテーションし、茎数という植物形質を推定する画像解析・機械学習手法の開発と評価が中心である。
abstractThis study aims to build a model to count the number of alfalfa stems from proximal images.
Accurate measurement of structural phenotypes, such as plant height and canopy width, is crucial for the scientific management of lettuce cultivation in Plant Factories with Artificial Lighting (PFALs). In this study, we developed a multimodal image fusion model using visible images (RGB), depth images (Depth), and infrared images (IR) to extract lettuce phenotypes in PFAL environments. We proposed a Residual Space information enhancement module (DRS) and a fusion feature supplement method with adaptive weight optimization for IR features (IRC) to address the weak spatial perception of traditional RGB-based models and the feature loss of RGB due to illumination disturbance. Three lettuce varieties (Bixiao, Huqian, and Mondai) were selected as experimental subjects to evaluate the robustness of our proposed model. In ablation experiments, the benchmark model improved by DRS increased by 1.6% and 0.9% in terms of MAP0.75 and MAP0.5:0.95, respectively. The benchmark model improved by IRC increased by 0.2%, 0.6%, and 1.2% in terms of MAP0.5, MAP0.75, and MAP0.5:0.95, respectively. Furthermore, MAP0.5, MAP0.75, and MAP0.5:0.95 values increased by 0.3%, 3.3%, and 2.3% when the two modules were combined, respectively. Compared with manually measured plant height and canopy width, the Root Mean Square Error (RMSE) of the average plant height prediction results for the three varieties is 0.74, and the Mean Squared Error (MSE) is 0.55. For canopy width, the RMSE of the model’s prediction results was 0.70, and the MSE was 0.49. In the lighting influence experiment, our method outperformed the unimproved model by approximately 0.3–4% in terms of MAP0.5, MAP0.75, and MAP0.5:0.95 across multiple datasets. Our proposed model effectively addresses lighting disturbance, enhances the robustness of the baseline model against varying lighting conditions, improves spatial perception capability, facilitates the separation of adjacent plant features in the model’s feature extraction stage, enhances the model’s detection ability, and ultimately improves phenotype extraction capability.
Why it matches plant phenotyping methodsマルチモーダル画像融合モデルを開発し、レタスの草丈・群落幅という植物表現型を画像から抽出・推定しており、手法開発と技術評価が研究の中心である。
abstractIn this study, we developed a multimodal image fusion model using visible images (RGB), depth images (Depth), and infrared images (IR) to extract lettuce phenotypes in PFAL environments.
With a rich historical background, the US cotton industry consistently maintains its position as one of the leading global producers. Due to the direct correlation between cottonseed quality and germination rate, conducting non-destructive testing to identify defects in cottonseeds becomes important to optimize yield performance. In this study, we propose an objective method for detecting cottonseed defects which classifies cottonseeds into four categories (Normal, Pinhole, Damage, and Very Damaged) and fourteen subcategories (N, R, C, RH, EH, CH, R Cut, C Cut, RV, CV, RH Expose, EH Expose, CH Expose, and V). Leveraging our customized cottonseed image dataset, we introduce a cottonseed defect detection and classification method based on a lightweight YOLOv5n model enhanced with Swin Transformer and an improved two-stage deep learning classification model. For cottonseed detection, our method achieves a 30.11 % reduction in model size and a 7.7 % increase in mAP50:95 compared to YOLOv5n. For individual cottonseed image classification, the accuracy, precision, recall, and F1 scores of our two-stage deep learning model are 97.34 %, 97.7 %, 97.3 %, and 97.3 %, respectively. The gradient-weighted class activation mapping (Grad-CAM) algorithm was then used to visually explain the model’s classification mechanism. Moreover, our algorithm demonstrates superior performance compared to six commonly used classification algorithms, including ResNet-18, ResNet-50, AlexNet, GoogleNet, VGG-16, and VGG-19, achieving a notable 1.65 % increase in accuracy over the best-performing algorithm among them. We then compared its performance with four state-of-the-art (SOTA) cottonseed damage classification methods. The findings demonstrate the potential for this design to advance the development of non-destructive seed damage detection.
Why it matches plant phenotyping methods綿実の損傷状態という植物器官の表現型を画像から分類・検出する深層学習手法を開発し、精度比較・検証しており、表現型取得・抽出法が研究の中心である。
abstractwe propose an objective method for detecting cottonseed defects which classifies cottonseeds into four categories
The aim of this paper is to address the lack of standard methodologies for the assessment of 3D point clouds. We present a methodology to realistically assess the accuracy of 3D point clouds, enabling the evaluation in a full 3D context rather than based on isolated points. Additionally, it introduces three significant innovations: a) it bridges the gap related to the unknown error of the reference ground-truth point cloud; b) it provides separate metrics for location error and reconstruction error; and c) it introduces a procedure to compute the location error that eliminates the bias in the selection of point-pair picking between the DGT points and their corresponding pairs in the point cloud being assessed. The geometry and structure of trees are related to the vegetative parameters and productivity in fruit orchards. In consequence, obtaining a precise and accurate geometric characterization of canopies is of interest for implementing site-specific management strategies that optimize input rates and minimize the costs and environmental risks of agricultural operations. Among the different sensing technologies, sensors based on the principle of light detection and ranging (LiDAR) have emerged as the primary choice for accurate geometric characterization of orchards. However, to make informed orchard management decisions based on LiDAR-derived geometric and structural data, it is essential to assess the accuracy of LiDAR-based scanning systems. Unfortunately, there is currently a lack of standard methodologies to evaluate the accuracy of LiDAR-based systems in agricultural environments. This research paper presents a novel methodology to assess the location error and the reconstruction error of 3D point clouds in full 3D context. The methodology involves comparing LiDAR-derived point clouds to an accurate high-resolution 3D digital ground truth (DGT) obtained using digital photogrammetric techniques. One of the main difficulties when using a reference point cloud to assess point cloud errors is the selection of the points to be compared so that they can be considered as corresponding point pairs. When developing the methodology, four procedures of point pair selection and distance calculation were compared. The best performing procedure was selected and proposed as a standard for accuracy assessment of 3D point clouds. The proposed procedure minimizes the error attributed to the selection of the corresponding point pairs between the assessed point cloud and the reference DGT point cloud. Subsequently, the proposed methodology was tested and validated by assessing the accuracy of 46 different point clouds. The conclusions regarding the accuracy, applicability, and practical utility of the proposed methodology are supported by the determination of reconstruction errors and location errors in 46 point clouds obtained with the 3 different MTLS systems operated with different settings. The proposed methodology will be very useful for scanning system manufacturers, researchers, advisors and eventually advanced farmers to quantify the errors committed when characterizing tree canopies. This is crucial to enable accurate management operations in the framework of Precision Agriculture based on canopy variability. Furthermore, the methodology is expected to facilitate the design of new applications requiring high accuracy to be implemented in the near future.
Why it matches plant phenotyping methods果樹キャノピーの3D形状・構造を測定するLiDAR点群について、誤差評価手法を開発し、46点群で検証しており、植物表現型取得の技術的評価が中心である。
abstractThis research paper presents a novel methodology to assess the location error and the reconstruction error of 3D point clouds in full 3D context.
Reliable, quantitative information on the presence and severity of crop diseases is essential for site-specific crop management and resistance breeding. Successful analysis of leaves under naturally variable lighting, presenting multiple disorders, and across phenological stages is a critical step towards high-throughput disease assessments directly in the field. Here, we present a dataset comprising 422 high resolution images of flattened leaves captured under variable outdoor lighting with polygon annotations of leaves, leaf necrosis and insect damage as well as point annotations of Septoria tritici blotch (STB) fruiting bodies (pycnidia) and rust pustules. Based on this dataset, we demonstrate the capability of deep learning for keypoint detection of pycnidia (F1=0.76) and rust pustules (F1=0.77) combined with semantic segmentation of leaves (IoU=0.96), leaf necrosis (IoU=0.77) and insect damage (IoU=0.69) to reliably detect and quantify the presence of STB, leaf rusts, and insect damage on symptom level under natural outdoor conditions. An analysis of intra- and inter-annotator agreement on selected images demonstrated that the proposed method achieved a performance close to that of annotators in the majority of the scenarios. We validated the generalization capabilities of the proposed method by testing it on images of unstructured canopies acquired directly in the field and without manual interaction with single leaves. This enables significantly higher throughput and automated data acquisition, which is critical to harness the full potential of image-based disease assessments. Model predictions were in good agreement with visual assessments of in-focus regions in these images, despite the presence of new challenges such as variable orientation of leaves and more complex lighting. This underscores the principle feasibility of diagnosing and quantifying the severity of foliar diseases under field conditions using the proposed imaging setup and image processing methods. By demonstrating the ability to diagnose and quantify the severity of multiple diseases in highly complex field scenarios, we lay the groundwork for high-throughput in-field assessments of foliar diseases that can support resistance breeding and the implementation of core principles of precision agriculture.
Why it matches plant phenotyping methods圃場画像から葉の壊死、病斑・病原体構造、害虫被害を検出・定量し、深層学習手法をデータセットで検証した研究であり、植物病害状態の表現型取得が中心です。
abstractHere, we present a dataset comprising 422 high resolution images of flattened leaves captured under variable outdoor lighting with polygon annotations of leaves, leaf necrosis and insect damage as well as point annotations of Septoria tritici blotch (STB) fruiting bodies (pycnidia) and rust pustules.
Rapeseed seedling growth monitoring indicates growth status and detects problems, such as seedling gaps, seedbed unevenness, and diseases or insect pests in time, which play an important role in improving sowing strategies, promoting the decision-making of fertilizer prescription, and increasing economic efficiency. To improve the accuracy of rapeseed seedling growth assessment, a multi-growth stage growth assessment method based on unmanned aerial vehicle (UAV) low-altitude remote sensing and semantic segmentation was proposed to assess the growth of rapeseed into excellent, average, and poor growth. First, to address the problem of complex field scenes and densely planted rapeseed leading to difficult segmentation of rapeseed seedlings and field drains, the original Deeplabv3+ model was improved by selecting the lightweight network MobileNetV2 as the backbone feature extraction network and fusing the coordinate attention（CA）module, which enables the model to better noise removal and feature extraction and improves the model’s accuracy and robustness. Then, a field drain optimal centerline algorithm is proposed to obtain the optimal centerline of all field drain in the image and determine the field box position. Finally, eight growth-related feature values for rapeseed seedling were constructed, and were used as feature vectors in a random forest (RF) to construct multi-growth stage growth assessment model for rapeseed seedlings. The results indicate that the improved DeeplabV3+ network outperformed the original DeeplabV3+ network, with the mean pixel accuracy increasing from 78.43 % to 87.47 % (an improvement of 9.04 %) and the average intersection over union (mIoU) increasing from 67.45 % to 76.89 % (an improvement of 9.44 %). The mean positional deviation of the centerline was –5.29 pixels with a standard deviation of 9.51 and a mean angular deviation of –0.01848 rad with a standard deviation of 0.00791, which can effectively detect the centerline of the field drain. The precision, sensitivity, specificity, and accuracy of the proposed method were 96.35 %, 96.34 %, 97.20 %, and 96.34 %, respectively. The algorithm of this study can efficiently segment rapeseed seedlings and field drains, obtain the optimal centerline of the field drain, and be used for rapeseed seedling multi-growth stage growth monitoring, which provides a theoretical basis and technical reference for rapeseed seedling multi-growth stage growth monitoring.
Why it matches plant phenotyping methodsUAV画像とセマンティックセグメンテーションを用いて、菜種苗の生育関連形質を抽出・評価する手法を開発し、精度検証まで行っており、フェノタイピング手法が中心である。
abstracta multi-growth stage growth assessment method based on unmanned aerial vehicle (UAV) low-altitude remote sensing and semantic segmentation was proposed
Potato (Solanum tuberosum) is widely recognized as the leading vegetable crop in the United States, with millions of tons produced annually. Despite many advancements in cultivars, crop production still suffers from meager progress in the assessment of early maturity. One potential solution to this problem is Ground-Penetrating Radar (GPR), a near-surface geophysical tool that has recently been applied to agriculture for assessment of root systems by detecting dielectric variations in sub-surface and soil layers by means of electromagnetic waves emitted into the ground. This study seeks to assess GPR’s capability to serve as a non-destructive proximal-sensing technique for quantifying potato tuber biomass by estimating the size of potatoes by measuring changes in the reflected GPR signal. Two methods, thresholding analysis and continuous wavelet transform (CWT), were employed in this study to extract features from GPR responses to predict tuber biomass. The dataset was collected in a controlled sandbox system. Thresholding analysis on the interpolated amplitude values yielded significant results, being able to predict tuber biomass with an accuracy of r = 0.82 and R2 = 0.64 based upon Multiple Linear regression. CWT was somewhat less successful, yet still significant, with a prediction accuracy of r = 0.6 and R2 = 0.32. These results indicate that GPR technology is suitable as a decision-support tool for potato breeders seeking to monitor tuber growth.
Why it matches plant phenotyping methodsGPRを用いてジャガイモ塊茎バイオマスを非破壊推定し、thresholdingとCWTによる特徴抽出・予測精度を評価しており、植物形質取得法が研究の中心です。
abstractThis study seeks to assess GPR’s capability to serve as a non-destructive proximal-sensing technique for quantifying potato tuber biomass
Stem / branchObject detectionGrowth / development / phenology
Machine vision plays a pivotal role in automatically monitoring the growth of mulberry branch cuttings (Mbc) in aeroponic systems, ensuring productivity, quality, and sustainability. However, the challenge lies in the varying bud growth rates, with some taking longer to break dormancy, leading to inconsistent development and delays in root formation. This paper aims to develop an Internet of Things (IoT)-integrated delta robot, equipped with advanced camera data acquisition and intelligent processing. It is intended to enhance aeroponic systems by precisely and rapidly detecting the Mbc growth state and applying growth stimulation. The system framework requires tiny machine learning models specifically designed to function efficiently on IoT hardware with limited power and resources, such as the lightweight versions of Tiny-YOLO, including YOLOv8-world, YOLOv9, YOLOv10, and YOLOv11. These models were trained on 3,000 images captured from three distinct camera perspectives—side view, elevated view, and angled view—during the growth of Mbc. Further optimization was achieved by progressively refining the weights through stepwise training. Artificial neural networks, for instance Back-Propagation Neural Networks (BPNN) and Elastic Net (ELNET), were deployed to compute the X, Y, and Z coordinates of the robot arm. Alongside, the look-up table approach efficiently identified the Mbc locations by referencing pre-stored data corresponding to the target coordinates. The experimental outcomes indicated that the optimized Tiny-YOLOv9 (Pr = 98.3 %, Re = 98.5 %, Fm = 98.4 %, mAP₅₀–₉₀ = 87.3 %) outperformed other models in both classification and localization of mulberry branches. Data fusion from three cameras with BPNN and ELNET models substantially outshined the use of a single camera. Moreover, the BPNN models for the arm axes (X, Y, Z) exhibited superior accuracy compared to ELNET. The BPNN recorded R² values of 0.999 for all three axes (X, Y, and Z), with corresponding RMSE values of 0.005 (MAE = 0.004), 0.006 (MAE = 0.005), and 0.013 (MAE = 0.010), respectively. This work can assist the agricultural community in monitoring plant growth, enabling timely and effective management decisions. It also holds great promise for expanding our methodology to include other crops in future aeroponic systems.
Why it matches plant phenotyping methods植物の成長状態をカメラ画像と機械学習で検出するIoT・ロボット型フェノタイピング基盤の開発が中心であり、単なる生物学的実験での routine 測定ではない。
abstractMachine vision plays a pivotal role in automatically monitoring the growth of mulberry branch cuttings (Mbc) in aeroponic systems
Grapevine phenotyping, that is the process of determining the physical properties (e.g., size, shape, and number) of grape bunches, provides valuable information for growth and health monitoring, yield estimation and efficient crop management in precision viticulture. Currently, grape bunch counting and sizing is done manually, which is labor intensive and often impractical for large-scale field applications. This paper describes a novel framework to automatically detect, count and estimate the volume/weight of grape bunches using RGB and depth data acquired in the field by a farmer robot. The proposed pipeline starts with the semantic segmentation of RGB images based on a pre-trained MANet architecture with EfficientnetB3 backbone to separate fruit from non-fruit regions. The segmented fruit mask is then projected onto the co-registered depth image to recover a depth mask, allowing for three-dimensional (3D) data association. After a pre-processing step to correct anomalies, such as corrupted and missing values, and to remove outliers, a depth gradient-based clustering algorithm is applied that detects individual grape bunch clusters. This enables the separation of adjacent and partially overlapping bunches. In addition, a method to reconstruct the whole 3D shape of a bunch is introduced, so as to provide an estimate of volume and weight. Experiments performed in a commercial vineyard in Italy are presented showing that, despite the low quality and high variability of the input images, the proposed approach is able to count grape bunch clusters with an average error of about 12% with respect to visual ground-truth and an average error less than 30% with respect to manual weight measurements. It is also shown that the processing framework can be applied to geo-referenced image sequences acquired by the farmer robot while traversing vineyard rows, thus providing an automated pipeline for the generation of high-resolution yield maps for precision viticulture applications.
Why it matches plant phenotyping methodsブドウ房の検出・計数・3D形状復元により体積・重量を推定する画像・深度ベースの表現型取得手法を開発し、精度検証も行っているため、方法が中心的である。
abstractThis paper describes a novel framework to automatically detect, count and estimate the volume/weight of grape bunches using RGB and depth data acquired in the field by a farmer robot.
The number of cotton bolls is an important phenotyping trait not only for breeders but also for growers. It can provide information on the physiological and genetic mechanisms of plant growth and aid decision-making in crop management. However, traditional visual inspection in the field is time-consuming and laborious. With the application of drones in the agricultural domain, there is promising potential to collect data expediently. In this paper, we integrated the improved Distribution Matching for crowd Counting (DM-Count) and Segment Anything Model (SAM) to predict cotton boll number, size, and yield in aerial images. The cotton plots were first extracted from the raw aerial images using boundaries derived from orthophotos. Then, a convolutional neural network (DM-Count) was introduced as a baseline and customized by replacing the VGG19 backbone and adding a pixel loss. The customized network was first pretrained on ground images and then fine-tuned on aerial images to predict the density map, where the number and locations of cotton bolls can be obtained. The zero-shot foundation model SAM was investigated to segment cotton bolls with the point prompts provided by customized DM-Count. The respective numbers of bolls and segmented pixels were compared for seed cotton yield estimation. The experimental results showed that the customized model obtained a mean absolute error (MAE) of 1.78 per square meter and a mean absolute percentage error (MAPE) of 4.39 % on the testing dataset, with a high correlation between the predicted boll number and ground truth (R² = 0.91). The AP50 of SAM for cotton boll segmentation was 0.63. The segmented masks were used to delineate the boll size differences among the four genotypes, and it was found that the average boll size of Pima was 452 pixels, which was significantly smaller than Acala Maxxa, UA 48 and Tamcot Sphinx. Moreover, the yield estimation using the boll number was better than that using the pixel number, with an R² = 0.70. Combing the boll number and the pixel number can achieve a slightly higher R² of 0.72 for yield estimation. Overall, the customized model can count cotton bolls in aerial images accurately and estimate seed cotton yield effectively, which could significantly benefit breeders in developing genotypes with high yields, as well as help growers in yield estimation and crop management.
Why it matches plant phenotyping methods綿花のボール数・サイズ・収量を航空画像から推定する画像解析手法を開発・評価しており、植物形質取得が研究の中心である。
abstractIn this paper, we integrated the improved Distribution Matching for crowd Counting (DM-Count) and Segment Anything Model (SAM) to predict cotton boll number, size, and yield in aerial images.
The effective tiller number of wheat (ETNW) is one of the three main factors affecting wheat yield. Most traditional methods for counting tillers are manual, which is inefficient and challenging to implement on a wide scale. Existing remote sensing techniques for tiller counting are typically focused on individual plants or small plot areas, often utilizing high-resolution sensors for close-range monitoring. This approach limits the scalability and applicability for large-scale field environments. Moreover, most studies on wheat tiller monitoring have concentrated on specific crop varieties or single-variable conditions, with little research conducted on estimating or monitoring ETNW in large-scale field scenarios using consumer-grade unmanned aerial vehicle (UAV) platforms. To address these issues, we propose a multi-modal fusion-driven machine learning method to enhance the performance of wheat tillering monitoring. This study was conducted using two experimental setups to ensure robust and comprehensive data collection. Experiment 1 (Exp.1) focused on water and nitrogen coupling conditions, while Experiment 2 (Exp.2) included both nitrogen-deficient and nitrogen-sufficient treatments. Each experiment involved multiple wheat varieties to account for genotypic variability. UAV data, including multispectral and RGB imagery, was collected across different growth stages under varying irrigation and nitrogen conditions to ensure the generalizability of the proposed model. This method employs spectral correlation analysis (SCA) to select features strongly correlated with the number of tillers and subsequently fuses these features to construct a multi-modal machine learning model based on vegetation index (VI), color index (CI), multispectral texture features (TF1), and RGB texture features (TF2) derived from UAV data. Three machine learning models are applied: Random Forest Regression (RFR), Partial Least Squares Regression (PLSR), and Support Vector Regression (SVR). The results demonstrated that the vegetation index-driven machine learning model (VI-RFR) achieved the best performance, with an R² of 0.85, RMSE of 348.60, and NRMSE of 0.367, followed by the color index model (CI-RFR, R² = 0.83, RMSE = 313.18, NRMSE = 0.330), the multispectral texture feature model (TF1-RFR, R² = 0.81, RMSE = 357.12, NRMSE = 0.376), and the RGB texture feature model (TF2-RFR, R² = 0.80, RMSE = 373.69, NRMSE = 0.393). Compared to using a single data source, data fusion significantly improved model accuracy, particularly when complementary data sources were combined. Specifically, the VI&CI combination achieved an R² improvement of 6.4 %-19.1 %, an RMSE reduction of 2.8 %-9%, and an NRMSE decrease of 1.56 %-8.24 %. The VI&TF1 combination exhibited an R² increase of 7.8 %–32.6 %, an RMSE reduction of 8.5 %-28.8 %, and an NRMSE decrease of 2.20 %-8.24 %. The VI&TF2 combination showed an R² increase of 3.8 %-40.5 %, an RMSE reduction of 2.8 %-27.6 %, and an NRMSE decrease of 2.68 %-21.61 %. The CI&TF1 combination resulted in an R² improvement of 5.2 %-25.6 %, an RMSE reduction of 7.4 %-24.1 %, and an NRMSE decrease of 2.04 %-19.41 %. The CI&TF2 combination achieved an R² increase of 12.5 %–33.3 %, an RMSE reduction of 5.4 %–23.4 %, and an NRMSE decrease of 1.17 %-18.94 %. The TF1&TF2 combination achieved an R² improvement of 13.9 %–33.3 %, an RMSE reduction of 10.9 %-27.6 %, and an NRMSE decrease of 5.01 %-21.61 %. However, increasing the number of data sources does not necessarily lead to higher model accuracy. The VI&CI&TF1 fusion model (RFR, R² = 0.90, RMSE = 288.39, NRMSE = 0.304) demonstrated the best performance, surpassing the combination of all four features. The machine learning models exhibited systematic variations, with the performance ranking as follows: RFR > SVR > PLSR, regardless of the data fusion strategy employed. Moreover, the best model (VI&CI&TF1-RFR) demonstrated adaptability across different growth stages, irrigation conditions, and nitrogen fertilizer applications. Notably, the model maintained robustness even in extreme environments with complete nitrogen deficiency. The findings of this study provide technical support for crop phenotyping, variety selection, and precision agriculture management.
Why it matches plant phenotyping methodsUAVのマルチスペクトル・RGB画像から小麦の有効分げつ数を推定する特徴融合・機械学習手法を開発し、複数条件・品種で性能評価しており、植物表現型の取得・推定が研究の中心である。
abstractTo address these issues, we propose a multi-modal fusion-driven machine learning method to enhance the performance of wheat tillering monitoring.
Phenotypic traits of crops reflect their physiological characteristics and provide a theoretical basis for predicting their growth. The 3D point cloud has a direct and accurate rendering ability, which has been widely used in phenotype extraction, especially with the help of accurate segmentation techniques. However, the inherent discrete nature of point clouds makes accurate organ segmentation an ongoing challenge in the field. In this study, we propose a tree phenotype acquisition method based on point cloud registration and skeleton segmentation. First, the Convex Hull-indexed Gaussian Mixture Model (CH-GMM) is employed to register the ground and aerial point cloud data. Then, a Laplace-multi-scale adaptive algorithm (LMSA) was proposed to obtain the crop skeleton structure, on the basis of which four phenotypic parameters, namely, plant height, crown width, branching number, and initial branching height, were extracted for fruit trees. In addition, the relationship between crown width and the number of branches was explored, where branches included initial, secondary, and tertiary branches. The results show that the proposed CH-GMM algorithm has a rotation error of less than 1.01°, a translation error of less than 10 mm, and a success rate of more than 95 %. The average precision, average recall, average F1 score, and average overall accuracy of the LMSA are 93.7 %, 96.2 %, 92.6 %, and 95.3 %, respectively. Finally, this study found a polynomial and exponential relationship between the number of bifurcations and crown size of fruit trees. The results of this study may provide new ideas for fruit tree phenotype acquisition and phenotype management.
Why it matches plant phenotyping methodsリンゴ樹の3D点群登録・骨格分割アルゴリズムを開発し、樹高や樹冠幅などの形態形質を抽出・精度評価しており、表現型取得手法が研究の中心である。
abstractwe propose a tree phenotype acquisition method based on point cloud registration and skeleton segmentation.
Extracting cotton boll phenotypic parameters from imaging data is a prerequisite for intelligently characterizing boll growth and development. However, current methods relying on manual measurements are inefficient and often inaccurate. To address this, we developed a cotton boll phenotypic parameter extraction program (CPVS), a tool designed to estimate the morphological characteristics of unopened cotton bolls from images. CPVS integrates semi-automatic data extraction with advanced algorithms to calculate length, width, volume, and surface area. Length and width estimation algorithms were developed using a custom “Fixed” image set, which links pixel dimensions to actual measurements. Volume and surface area models were based on shape classification using a custom “Random” image set, trait correlations, and measured data. Testing showed strong performance, with R² values of 0.880 and 0.769 and root mean square error (RMSE) values of 0.173 and 0.188 for length and width, respectively. The volume model achieved an R² of 0.91 and an RMSE of 1.76, while surface area models had R² values of 0.76 and RMSEs of 2.37 and 2.41. These results indicate that CPVS is a robust tool, providing theoretical and practical support for efficient, accurate characterization of cotton boll morphology.
Why it matches plant phenotyping methods画像から未開裂綿花果の形態形質を抽出するソフトウェアと推定モデルを開発・検証しており、植物表現型取得が研究の中心です。
abstractwe developed a cotton boll phenotypic parameter extraction program (CPVS), a tool designed to estimate the morphological characteristics of unopened cotton bolls from images.
Accurate and precise spraying in orchards is paramount for optimized agricultural practices, ensuring efficient pesticide utilization, minimized environmental impact, and enhanced crop yield by targeting specific areas with the right amount of treatment. The asymmetrical distribution of foliage and flowers in peach orchards poses a formidable challenge to achieving precise spray accuracy, impeding the uniform application of treatments and compromising the overall efficacy of pest and disease control measures. In response to the prevailing challenges in achieving accurate spray application caused by the asymmetrical distribution of foliage and flowers in peach orchards, this paper introduces a novel deep neural network to map the RGB image and corresponding depth to the density map of peach flowers or foliage. The model consists of components: (1) two backbones based on ResNet-50 that extract contextual features from the RGB image and depth features from depth data at multiple scales and levels; (2) an optimized depth-enhanced module that effectively fuses the distinct features extracted from the two input streams; and (3) a two-stage decoder that aggregates the high-level cross-modal features to regress the coarse density map and subsequently integrates it with the low-level cross-modal features for final density map prediction. To evaluate the performance of our model, we collected 493 frames (206,095 instances) of peach flowers and 475 frames (350,833 instances) of foliage from the peach orchards utilizing our sprayer prototype equipped with stereo cameras. The proposed method outperforms state-of-the-art models on our datasets, demonstrating the superiority and efficacy for encoding canopy characteristics in the form of flower and foliage density maps for blossom and cover sprays. It attains significant computational efficiency, exhibiting a frame rate of 20 FPS, and showcases exceptional accuracy with a WMAPE of 12.11% for peach flowers and a WMAPE of 13.37% for leaves.
Why it matches plant phenotyping methods桃の花・葉の密度という植物器官の形質をRGB-D画像から推定する手法を開発・評価しており、散布制御への応用を超えてフェノタイピング手法自体が中心です。
abstractthis paper introduces a novel deep neural network to map the RGB image and corresponding depth to the density map of peach flowers or foliage
Monitoring crop N status by means of proximal and remote sensing data can help enhancing N use efficiency at various farm scales. This study compares five optical sensor platforms, commonly used in practice and research, based on their usability and accuracy in measuring crop N status at field level. The data were gathered in 2019 in two sites in northeast Switzerland that were cropped with winter wheat (Triticum aestivum). The optical sensor platforms employed included a Sentinel-2 satellite, two different unmanned aircraft systems (UAS fixed-wing and quadcopter), a tractor-mounted system, and a handheld field spectrometer. We used a power regression to compare the measured crop N uptake with spectral vegetation indices computed from the different sensors. The reported normalized difference red-edge (NDRE) index values were distributed in a broad range from 0.17 to 0.74, with the Sentinel-2 satellite records in the higher part of the range (0.59–0.74) and those of the handheld spectrometer in the low range (0.17–0.29). The study’s key finding was the information collected was significantly different across the five sensing platforms, in terms absolute values from the sensors. However, the correlations between NDRE values from all sensors and the measured N uptake were comparably robust, with r > 0.8 a root mean square error ranging from 29 to 37 kg N/ha. Furthermore, the N application maps produced for the satellite and UAS platforms showed that the best compromise between detailed spatial resolution and matching of the working width of the machinery used was achieved by resampling the UAS-based maps at 10 m resolution with the calculation used in this study. We concluded that sensor-based N status assessment across different sensing levels can support the improvement of N use efficiency by allowing a more precise management of in-field variability, with the precondition of having a good calibration for climatic location and variety. However, factors such as the degree of detail needed to capture in-field variability while matching the working width should be evaluated for each specific case.
Why it matches plant phenotyping methods冬コムギのN状態を推定する5種類の光学センシング基盤を比較し、精度・相関・RMSE・空間解像度を評価しているため、植物生理状態の取得方法が研究の中心です。
abstractThis study compares five optical sensor platforms, commonly used in practice and research, based on their usability and accuracy in measuring crop N status at field level.
Accurate estimation of above-ground biomass (AGB) in potato plants is essential for effective monitoring of potato growth and reliable yield prediction. Remote sensing technology has emerged as a promising method for monitoring crop growth parameters due to its high throughput, non-destructive nature, and rapid acquisition of information. However, the sensitivity of remote sensing vegetation indices to crop AGB parameters declines at moderate to high crop coverage, known as the “saturation phenomenon,” which limits accurate AGB monitoring during the mid-to-late growth stages. This challenge also hinders the development of a multi-growth-cycle AGB estimation model. In this study, a novel VGC-AGB model integrated with hyperspectral remote sensing was utilized for multi-stage estimation of potato AGB. This study consists of three main components: (1) addressing the “saturation problem” encountered when using spectral indices from remote sensing to monitor crop biomass across multiple growth stages. The VGC-AGB model calculates the leaf biomass by multiplying leaf dry mass content (Cm) and leaf area index (LAI) and vertical organ biomass using the multiplication of crop density (Cd), crop height (Ch) and the crop stem and reproductive organs’ average dry mass content (Csm); (2) estimating the VGC-AGB model parameters Cm and LAI by integrating hyperspectral remote sensing data with a deep learning model; (3) comparing the performance of three methods—(i) hyperspectral + Ch, (ii) ground-measured parameters + VGC-AGB model, and (iii) hyperspectral remote sensing + VGC-AGB model—using a five-year dataset of potato above-ground biomass. Results indicate that (1) the VGC-AGB model achieved high accuracy in estimating AGB (R² = 0.853, RMSE = 751.12 kg/ha), significantly outperforming the deep learning model based on hyperspectral + Ch data (R² = 0.683, RMSE = 1122.03 kg/ha); (2) the combination of the VGC-AGB model and hyperspectral remote sensing provided highly accurate results in estimating AGB (R² = 0.760, RMSE = 965.59 kg/ha), surpassing the results obtained using the hyperspectral + Ch-based method. Future research will primarily focus on streamlining the acquisition of VGC-AGB model parameters, optimizing the acquisition and processing of remote sensing data, and enhancing model validation and application. Furthermore, it is essential to conduct cross-regional validation and optimize model parameters for various crops to improve the universality and adaptability of the proposed model.
Why it matches plant phenotyping methodsジャガイモの地上部バイオマスという植物形質を、ハイパースペクトルリモートセンシングと深層学習・VGC-AGBモデルで推定し、複数手法と5年データで性能比較しているため、形質取得・推定手法が中心である。
abstracta novel VGC-AGB model integrated with hyperspectral remote sensing was utilized for multi-stage estimation of potato AGB.
Rice blast is one of the most destructive diseases affecting rice crops globally, known for its rapid spread and significant damage. The initial phenotypic symptoms are challenging to identify, and conventional field detection methods frequently do not fulfill the necessary detection requirements. We have developed an innovative detection method for the symptomless stage of rice blast disease, utilizing metal oxide semiconductor (MOS) gas sensors in conjunction with multimodal data optimization techniques. The distinctive fingerprint characteristics of volatile organic compounds (VOCs) emitted by rice were systematically collected. Principal Component Analysis (PCA) was utilized to identify the relevant feature sensors. A significant correlation was observed between the responses of the sensors and the physiological indicators of rice leaves, including chlorophyll and hydrogen peroxide concentrations, with a correlation coefficient of r = -0.91. The criteria importance through intercriteria correlation (CRITIC) weighting method, along with the gaussian membership function (GMF), has been employed to analyze and integrate sensor feature data, thereby effectively differentiating the characteristics associated with the symptomless stage of rice blast. The processed features were utilized to train, calibrate, and validate a model based on Support Vector Machine (SVM) technology, culminating in the development of the CRITIC-GMF-SVM framework for the precise detection of rice blast during its symptomless stage. This model achieved an impressive accuracy rate of 98.37 % for the symptomless stage, significantly outperforming conventional on-site detection techniques. This method offers significant advantages in the early detection and precise monitoring of rice blast disease. Moreover, its low cost and high stability enable a wide range of applications, making it well-suited for complex field environments.
Why it matches plant phenotyping methodsイネの無症状病害という植物状態をMOSガスセンサーとデータ解析で検出する方法を開発し、学習・較正・検証しているため、植物フェノタイピング手法が中心である。
abstractWe have developed an innovative detection method for the symptomless stage of rice blast disease, utilizing metal oxide semiconductor (MOS) gas sensors in conjunction with multimodal data optimization techniques.
To investigate the dynamic changes in superoxide dismutase (SOD), peroxidase (POD), and catalase (CAT) activities in Tomato yellow leaf curl virus (TYLCV)-infected tomato leaves is essential for monitoring tomato growth, selecting disease-resistant varieties and disease control. Utilizing hyperspectral technique offers a rapid and non-destructive method to estimate antioxidant enzyme activity in tomato leaves. This study focuses on one a TYLCV-susceptible tomato variety (Hezuo 908) and two TYLCV-resistant varieties (Dingyanfen No. 3 and No. 5) to analyze the changes in photosynthetic characteristics and antioxidant enzyme activity following TYLCV infection. Hyperspectral data were used to quantify the correlation between antioxidant enzyme activity and spectral features under different pretreatments, as well as the relationship between enzyme activity and classical spectral indices. Several optimized indices were developed by iterating on the condition of R, forming a combined index. Considering efficiency and complexity, the support vector machine regression algorithm was employed to evaluate the predictive performance of the models. The results showed a decline in photosynthetic rate and relative chlorophyll content, while stomatal conductance initially decreased and then increased. The activities of the three antioxidant enzymes increased. activities of the three antioxidant enzymes post-TYLCV infection, with POD activity correlating with tomato variety resistance—a potential auxiliary index for antiviral variety identification. Among the models, the CAT prediction model performed the best, with a test set determination coefficient (R²) of 0.82, followed by POD with an R² of 0.67, and SOD R² of 0.43. These findings demonstrate that spectra enable rapid and non-destructive estimation of antioxidant enzyme activity in tomatoes under viral stress. This study provides valuable insights for antiviral variety breeding and the early warning of viral diseases.
Why it matches plant phenotyping methodsトマト葉の抗酸化酵素活性をハイパースペクトルデータから非破壊推定する手法を開発・評価しており、植物の生理状態・ウイルスストレスの表現型取得が中心である。
abstractUtilizing hyperspectral technique offers a rapid and non-destructive method to estimate antioxidant enzyme activity in tomato leaves.
The performance of neural networks is heavily dependent on the integrity of the feature signals. As these signals are extracted and transmitted, they tend to weaken, which can negatively affect their ability to represent and utilize semantic information, particularly in weakly supervised learning tasks. This study aims to address hidden and severely weakened signals. To address the underlying causes, the rectification block of the third stage of PR-ArsenicNetPlus (Positive Rectified ArsenicNetPlus Neural Network) was modified to include a nonlinear fitting method based on the variant Hölder inequality. This method adjusts the magnitude and angular frequency of an extremely weak signal, and its effectiveness is evaluated using Parseval’s relationship. When tested on cassava leaf disease datasets, the proposed method significantly improved the prediction accuracy in 7-fold cross-validation, achieving an accuracy of 96.18 %, a loss of 1.373, and an F1-score of 0.9618. These results outperformed those of ResNet-101, EfficientNet-B5, RepVGG-B3g4, and AlexNet.
Why it matches plant phenotyping methodsキャッサバ葉の病害状態を画像データから推定するニューラルネットワーク手法を開発し、既存モデルとの性能比較で検証しているため、植物フェノタイピング手法が中心である。
titleRectifying the extremely weakened signals for cassava leaf disease detection
Due to the wide variety of lotus species and the need for phenotypic categorization, traditional recognition is limited by the current manual observation and measurement of lotus phenotypes. In this paper, a lotus species recognition technique based on MobileNetV2-SE with reliable pseudo-labelling is proposed to construct an image dataset containing 94 different lotus species, and various data enhancement techniques are employed. Within MobileNetV2-SE, the classical MobileNetV2 network is improved by embedding the SE (Squeeze-and-Excitation) module, and then pseudo-labelling technical of semi-supervised learning is adopted to improve the classification performance of the model by generating high-quality labelling data. The test results show that the model in this paper can achieve an accuracy of 98.11% for lotus phenotype classification, and the precision, recall and F1 value can reach 98.45%, 98.47% and 98.40%, respectively, and the number of parameters and the amount of computation are 2.41×106 and 3.41×108 FLOPs, which are significantly better than other networks. This paper provides an effective solution for the automatic identification of lotus varieties and provides a reference for other plant variety identification tasks.
Why it matches plant phenotyping methods画像ベースでハスの表現型・品種を自動分類する深層学習手法を開発し、データセット構築と性能評価を行っており、表現型取得・分類手法が研究の中心である。
abstracta lotus species recognition technique based on MobileNetV2-SE with reliable pseudo-labelling is proposed to construct an image dataset containing 94 different lotus species
Timely and accurate prediction of nitrogen (N) status in winter wheat is crucial for guiding precision N management. This study aimed to develop an efficient model for predicting winter wheat plant N concentration (PNC) by integrating proximal hyperspectral sensing data with weather information. Hyperspectral data were collected from six field experiments conducted from 2014 to 2023, which were preprocessed using first-order derivative, log-transformation, and continuum removal methods. Effective spectral bands were selected by least absolute shrinkage and selection operator (LASSO), combined with weather information and analyzed using seven machine learning algorithms. The results indicated that first-order derivative-preprocessed bands combined with Elastic Net Regression provided the best PNC prediction (coefficient of determination (R²) = 0.78, root mean square error (RMSE) = 0.28 % and relative prediction deviation (RPD) = 2.15) among the tested methods. Combining proximal hyperspectral sensing and weather information with machine learning algorithms significantly enhanced winter wheat PNC predictions (R² = 0.79–0.85, RMSE = 0.23–0.27 % and RPD = 2.15–2.56) compared with using proximal hyperspectral sensing (R² = 0.34–0.79, RMSE = 0.28–0.48 % and RPD = 1.23–2.15) alone. This approach offers a promising framework for winter wheat PNC prediction to support precision N management. Future work should focus on developing multi-source data fusion strategies, incorporating unmanned aerial vehicle or satellite hyperspectral sensing and machine learning, for large-scale monitoring of crop N status and N management decision making.
Why it matches plant phenotyping methods近接ハイパースペクトルセンシングと機械学習により冬コムギの植物体窒素濃度を予測する手法を開発・比較評価しており、植物の生理状態の取得が研究の中心である。
abstractThis study aimed to develop an efficient model for predicting winter wheat plant N concentration (PNC) by integrating proximal hyperspectral sensing data with weather information.
Accurate detection and timely management of grapevine diseases, such as downy and powdery mildew, are essential for ensuring vineyard health and maximizing yield and quality. This study presents a novel approach using a ResNet50 model enhanced with batch normalization for precise identification and classification of grapevine leaf and fruit diseases. The dataset consists of 1,226 images categorized into five classes: Downy mildew diseased fruits (70 training, 18 testing), Downy mildew diseased leaves (534 training, 134 testing), Healthy leaves (183 training, 46 testing), Powdery mildew diseased fruits (93 training, 23 testing), and Powdery mildew diseased leaves (100 training, 25 testing). The model achieved an impressive accuracy of 95% in distinguishing between healthy and diseased grapevine leaves during rigorous evaluation and validation phases. Evaluation metrics—precision (94%), recall (96%), and F₁-score (95%)—highlight the model’s effectiveness compared to conventional methods. This research demonstrates the feasibility and superiority of deep learning in vineyard disease management, emphasizing its potential to revolutionize viticulture practices through automated, real-time disease detection and monitoring. These findings contribute to advancing agricultural sustainability and productivity through innovative technology applications in plant pathology.
Why it matches plant phenotyping methodsブドウ葉・果実の病害状態を画像から分類する深層学習手法を開発・評価しており、植物病害表現型の取得が研究の中心である。
abstractThis study presents a novel approach using a ResNet50 model enhanced with batch normalization for precise identification and classification of grapevine leaf and fruit diseases.
Yield prediction of root-fruit crops before harvest is significant for implementing precise field management. However, unlike crops such as wheat and corn, non-destructively predicting the yield of root-fruit crops non-destructively is challenging owing to their edible parts being located underground. Remote sensing offers a potential solution to this problem. Studies on predicting peanut yield through remote sensing are rare. Most of these studies relying on specific vegetation indices, such as the normalized difference vegetation index (NDVI), but have limitations in terms of model accuracy when other phenological parameters influencing peanut yield formation are not considered. 355 peanut yield samples were collected from two distinct cultivation patterns in 2022, 2023 and 2024 and In the study of peanut yield prediction, two modeling methods, linear regression and random forest, were employed to develop prediction models. Considering the contributions of early-stage material accumulation and late- stage material transfer to peanut yield, the results showed that incorporating multiple phenological parameters into peanut yield prediction models enhances accuracy beyond that achieved by models relying solely on early growth stage vegetation indices such as maximum NDVI.. Furthermore, the random forest algorithm has demonstrated its effectiveness in predicting peanut yields, particularly for summer peanuts, as evidenced by its successful application in related studies. The R² reached a high of 0.8201, while the lowest MAE and RMSE values were recorded at 0.2878 and 0.4048 t/ha, respectively. This study’s findings have significantly contributed to remote sensing-based yield prediction for root-fruit crops, further refining precision management practices in the cultivating of crops such as peanuts.
Why it matches plant phenotyping methodsリモートセンシングと回帰・ランダムフォレストによるピーナッツ収量推定モデルの開発が中心で、植物形質（収量）を定量化している。
abstracttwo modeling methods, linear regression and random forest, were employed to develop prediction models.
Citizen science is an effective approach for collecting extensive data scalable for deep learning, although data quality is debatable. However, few studies have determined the factors associated with data collection that affect model performance and potential sampling bias. This study aims to identify the factors that significantly influence the performance of a deep learning object detection model in agricultural prediction tasks. To do so, we analyzed errors in a You Only Look Once (YOLO v8) model trained for counting the number of coffee cherries in mobile pictures. The model was trained with 436 images taken in Colombia and Peru collected by local farmers as a citizen science approach. We analyzed the prediction errors of the model using 637 additional pictures. We then applied a linear mixed model (LMM) and a decision tree machine learning model to regress the model’s error against predictor variables related to the following categories: photographer influence, geographic location, mobile phone characteristics, picture characteristics, and coffee varieties. Our results show the strong influence of photographer identity and adherence (whether the image collection protocol was followed or not) on model prediction error. Following the protocol can increase model performance from an R2 of 0.48 to 0.73. Additionally, model performance varied significantly depending on photographer identity, with R2 ranging from 0.45 to 0.93. In contrast, factors such as mobile phone characteristics (e.g., frontal camera resolution, flash type, and screen size), using the screen behind the branch to obscure other cherries, coffee varieties, and geographic location did not significantly affect prediction error. These findings demonstrate that data quality in citizen science–based data collection for enhancing model prediction can be achieved through straightforward and comprehensive protocols, customized volunteer training, and regular feedback from experts. Such measures collectively support the robust application of deep learning models in agriculture. Furthermore, this study demonstrated that any mobile device with a camera can contribute to citizen science initiatives, underscoring the potential and scalability of this approach in agricultural research.
Why it matches plant phenotyping methodsコーヒーチェリー数という植物器官の形質を画像から推定する深層学習モデルを対象に、予測誤差、撮影者、プロトコル遵守などを分析して技術性能を検証しており、フェノタイピング手法が中心である。
abstractWe analyzed the prediction errors of the model using 637 additional pictures.
In recent years, sustainable agriculture has become increasingly important due to challenges such as climate change, population growth, and the need for food security. Tomato plants, being highly susceptible to various diseases, require accurate and timely diagnosis to maintain crop quality. Deep learning, particularly convolutional neural networks (CNNs), has shown great potential in addressing this challenge. This study introduces an advanced method for identifying tomato diseases using the CustomBottleneck-VGGNet model, enhanced through transfer learning techniques. The objective is to develop a highly accurate and computationally efficient model that can be deployed on resource-constrained devices for real-time disease detection. The proposed model achieves a remarkable accuracy of 99.12% with just 1.4 million parameters, outperforming classical models such as MobileNetV2, ResNet50, GoogleNet, VGG16, and VGG19 in terms of accuracy, precision, recall, and F1-score. Additionally, a mobile application has been developed to deploy this model, enabling real-time disease detection using a smartphone camera or images from the gallery, even in offline environments. The study also introduces a novel method for model comparison, focusing on differences between models trained under identical conditions to ensure fair evaluations.
Why it matches plant phenotyping methodsトマト葉の病害状態を画像から識別する深層学習モデルを開発・比較し、スマートフォン展開まで行っており、植物病害フェノタイピング手法が中心である。
abstractThis study introduces an advanced method for identifying tomato diseases using the CustomBottleneck-VGGNet model, enhanced through transfer learning techniques.
In this study, novel methods using portable NIR and Raman spectroscopy instruments associated with multivariate classification were developed to classify cotton fibers according to their length. The Upper Half Mean (UHM) length is considered a quality parameter by the cotton fiber market and is traditionally determined using a high-volume system (HVI), which entails high installation costs and labor-intensive analyses. As UHM correlates with cellulose polymerization, its determination can be achieved through vibrational spectroscopy techniques such as near-infrared (NIR) and Raman. These technologies offer advantages such as low cost, ease of handling, and rapid data acquisition, making them suitable for field use. This study aimed to develop a method and demonstrate the feasibility of using portable NIR and Raman spectrometers coupled with pattern recognition (PR) methods for routine analysis of cotton fibers, serving as a proof of concept for practical application in the industry. A total of 142 samples of cotton fibers from cotton improvement experiments conducted by the Brazilian Agricultural Research Corporation (EMBRAPA) were employed. Two classification approaches based on cotton lint length and the related economic value were employed. The first aimed to differentiate between short (SM) and long (LF) fibers, while the second aimed to further classify long fibers into internal classes (L, VL, and EL). Overall, methods using portable Raman spectrometer exhibited 100% accuracy performance regardless of the PR technique used. Meanwhile, methods based on NIR spectrometers achieved accuracies of 100% depending on the PR method and variable selection employed. The use of GLSW resulted in a reduction of a latent variable. In conclusion, the use of portable NIR and Raman spectrometers combined with PR methods emerges as an innovative and viable technology for the classification of cotton fibers based on their length.
Why it matches plant phenotyping methods携帯型NIR・ラマン分光と多変量解析を用いて、綿繊維長という植物由来形質を分類する手法を開発・実証しており、測定手法が研究の中心である。
abstractnovel methods using portable NIR and Raman spectroscopy instruments associated with multivariate classification were developed to classify cotton fibers according to their length.