Abstract Background Automating stomatal trait measurement has gained popularity because of their inherent importance for field phenotyping application as stomata are critical for both carbon capture and water use efficiency in plants. Such tool has been reported for rice, wheat, tomato, barley and oil palm. However, none exist yet for canola, which is an important economic and agronomic crop globally. Results We developed a new toolkit called Stomatal Comprehensive Automated Neural Network or SCAN by combining the use of high-resolution portable digital microscopy with machine learning based on You Only Look Once algorithm (YOLOv8). Digital micrographs of leaf surfaces enter the SCAN pipeline, which includes stomata detection, stomata segmentation and stomatal pore segmentation models, to measure stomatal density, stomatal size and stomatal pore area, respectively. In addition to SCAN’s ability to measure leaf stomatal traits in canola at 89 to 94% accuracy, we also showed that SCAN can be used to predict stomatal density even in species not included in the training set such as Arabidopsis, tobacco, rice, wheat, maize and proso millet. SCAN was designed for the biological science community with the premise that users are not required to possess advanced programming capabilities to manage dependency prerequisites, execute the models, and integrate the analysis. This was achieved by packaging the models into a desktop application system that can be accessed offline. Conclusion Overall, SCAN provides a non-destructive, real-time, portable, and high-throughput measurement of leaf stomatal traits in canola. The minimised hardware requirement and user-friendly desktop application system make SCAN suitable for field phenotyping application.
Why it matches plant phenotyping methodsカノーラ葉の気孔形質を画像と機械学習で自動抽出するツールを開発し、精度検証と他種での適用性評価を行っており、フェノタイピング手法が中心である。
abstractWe developed a new toolkit called Stomatal Comprehensive Automated Neural Network or SCAN by combining the use of high-resolution portable digital microscopy with machine learning based on You Only Look Once algorithm (YOLOv8).
Reproduction assets foundThe paper's authors publicly deposit the SCAN pipeline's model weights, hyperparameters, training scripts, and datasets in the FD_detection GitHub repository, and provide the SCAN application itself (with download and demonstration) in a second GitHub repository. Both are paper-specific, public, and actionable.Code · publicin Table S1.
123
124
The training tasks were carried out on an Ubuntu 20.04 Linux server at the Research School of Biology in
125
Australian National University, using two Nvidia A30 (24G) Graphic Processing Units (GPUs). The full
126
details of models’ weights, hyperparameters, training scripts and datasets can be found at
127
https://github.com/William-Yao0993/FD_detection.128
129
Model evaluation
130
131
Mean Average Precision (mAP, Fig. 3) and F1 score were used to assess model ability (Fig. 4). mAP is
132
calculated as the mean value of each class area under the precision-recall curve over thresholds, and the
133
F1 score is the harmonic mean of precision and recall. The formulas are deOpen asset ↗William-Yao0993/FD_detectionpdf-raw-page:4 lines:1-81Code / dataset availability confirmedarXiv · checked 15 Sept 2026
Providing an accurate evaluation of palm tree plantation in a large region can bring meaningful impacts in both economic and ecological aspects. However, the enormous spatial scale and the variety of geological features across regions has made it a grand challenge with limited solutions based on manual human monitoring efforts. Although deep learning based algorithms have demonstrated potential in forming an automated approach in recent years, the labelling efforts needed for covering different features in different regions largely constrain its effectiveness in large-scale problems. In this paper, we propose a novel domain adaptive oil palm tree detection method, i.e., a Multi-level Attention Domain Adaptation Network (MADAN) to reap cross-regional oil palm tree counting and detection. MADAN consists of 4 procedures: First, we adopted a batch-instance normalization network (BIN) based feature extractor for improving the generalization ability of the model, integrating batch normalization and instance normalization. Second, we embedded a multi-level attention mechanism (MLA) into our architecture for enhancing the transferability, including a feature level attention and an entropy level attention. Then we designed a minimum entropy regularization (MER) to increase the confidence of the classifier predictions through assigning the entropy level attention value to the entropy penalty. Finally, we employed a sliding window-based prediction and an IOU based post-processing approach to attain the final detection results. We conducted comprehensive ablation experiments using three different satellite images of large-scale oil palm plantation area with six transfer tasks. MADAN improves the detection accuracy by 14.98% in terms of average F1-score compared with the Baseline method (without DA), and performs 3.55%-14.49% better than existing domain adaptation methods.
Why it matches plant phenotyping methods油ヤシ個体の計数・検出という植物形態／個体数形質を衛星画像から推定する手法を開発し、アブレーション実験と既存手法比較で検証しており、フェノタイピング手法が中心である。
abstractwe propose a novel domain adaptive oil palm tree detection method, i.e., a Multi-level Attention Domain Adaptation Network (MADAN) to reap cross-regional oil palm tree counting and detection.
Reproduction assets foundThe authors explicitly state that their code and datasets (satellite images and annotations used for oil palm tree detection) are publicly available on GitHub.Code · publiche oil palm tree detection performance across different
remotely sensed images acquired from different sensors, regions and dates, without using labeled samples in the
target region. Our MADAN is proposed for enhancing both the generalization capacity and the transferability of
our model. Our codes and datasets are available on https://github.com/rs-dl/MADAN. The major contributions of
our work are as follows:
(1) We propose an adaptive object detector named MADAN for oil palm tree counting and detection across different
satellite images, which is the first work for large-scale domain adaptive tree crown detection using multi-source and
multi-temporal remote sensing images.
(2) WeOpen asset ↗rs-dl/MADANpdf-raw-page:6 lines:1-22