Grapevine yellows (GY) are a significant threat to grapes due to the severe symptoms and lack of treatments. Conventional diagnosis of the phytoplasmas associated to GYs relies on symptom identification, due to sensitivity limits of diagnostic tools (e.g. real time PCR) in asymptomatic vines, where the low concentration of the pathogen or its erratic distribution can lead to a high rate of false-negatives. GY’s primary symptoms are leaf discoloration and irregular wood ripening, which can be easily confused for symptoms of other diseases making recognition a difficult task. Herein, we present a novel system, utilizing convolutional neural networks, for end-to-end detection of GY in red grape vine (cv. Sangiovese), using color images of leaf clippings. The diagnostic test detailed in this work does not require the user to be an expert at identifying GY. Data augmentation strategies make the system robust to alignment errors during data capture. When applied to the task of recognizing GY from digital images of leaf clippings—amongst many other diseases and a healthy control—the system has a sensitivity of 98.96% and a specificity of 99.40%. Deep learning has 35.97% and 9.88% better predictive value (PPV) when recognizing GY from sight, than a baseline system without deep learning and trained humans respectively. We evaluate six neural network architectures: AlexNet, GoogLeNet, Inception v3, ResNet-50, ResNet-101 and SqueezeNet. We find ResNet-50 to be the best compromise of accuracy and training cost. The trained neural networks, code to reproduce the experiments, and data of leaf clipping images are available on the internet. This work will advance the frontier of GY detection by improving detection speed, enabling a more effective response to the disease.
Why it matches plant phenotyping methodsブドウ葉の画像から植物体の病徴（黄化病）をCNNで検出する手法を開発・評価しており、病害状態の取得が研究の中心であるため。
abstractwe present a novel system, utilizing convolutional neural networks, for end-to-end detection of GY in red grape vine (cv. Sangiovese), using color images of leaf clippings.
Reproduction assets foundThe authors explicitly state that the trained neural networks, code to reproduce the experiments, and the leaf clipping image dataset are publicly available on GitHub (Salento-Grapevine-Yellows-Dataset repository).Dataset · publicThe
trained neural networks, code to reproduce the experiments, and data of leaf clipping images are available on the
internet.Open asset ↗pdf-raw-page:1 lines:1-82Code · publicCode is publicly available on GitHub.Open asset ↗pdf-raw-page:7 lines:1-75Code / dataset availability confirmedEurope PMC · checked 15 Sept 2026
This paper provides synthesis methods for large-scale semantic image segmentation datasets of agricultural scenes with the objective to bridge the gap between state-of-the art computer vision performance and that of computer vision in the agricultural robotics domain. We propose a novel methodology to generate renders of random meshes of plants based on empirical measurements, including the automated generation per-pixel class and depth labels for multiple plant parts. A running example is given of Capsicum annuum (sweet or bell pepper) in a high-tech greenhouse. A synthetic dataset of 10,500 images was rendered through Blender, using scenes with 42 procedurally generated plant models with randomised plant parameters. These parameters were based on 21 empirically measured plant properties at 115 positions on 15 plant stems. Fruit models were obtained by 3D scanning and plant part textures were gathered photographically. As reference dataset for modelling and evaluate segmentation performance, 750 empirical images of 50 plants were collected in a greenhouse from multiple angles and distances using image acquisition hardware of a sweet pepper harvest robot prototype. We hypothesised high similarity between synthetic images and empirical images, which we showed by analysing and comparing both sets qualitatively and quantitatively. The sets and models are publicly released with the intention to allow performance comparisons between agricultural computer vision methods, to obtain feedback for modelling improvements and to gain further validations on usability of synthetic bootstrapping and empirical fine-tuning. Finally, we provide a brief perspective on our hypothesis that related synthetic dataset bootstrapping and empirical fine-tuning can be used for improved learning.
Why it matches plant phenotyping methods植物部位のセマンティックセグメンテーション用の合成・実画像データセットと生成手法を開発し、性能比較・検証可能な形で公開しており、植物画像から部位を抽出する方法が中心である。
abstractWe propose a novel methodology to generate renders of random meshes of plants based on empirical measurements, including the automated generation per-pixel class and depth labels for multiple plant parts.
Reproduction assets foundThe paper publicly releases its synthetic and empirical Capsicum annuum image datasets (with annotations) via a 4TU/Centre DOI, explicitly stated in the conclusion.Dataset · publicur experiments. Segmentation results
show a promising next step for semantic part localisation in agriculture.
Future efforts should be aimed in further optimising the network ar-
chitectures, focussing on the performance of the infrequent classes. The
datasets and their source material are publicly released and can be
found at: https://doi.org/10.4121/uuid:884958f5-b868-46e1-b3d8-a0b5d91b02c0
Acknowledgements
This research was partially funded by the European Commission in
the Horizon2020 Programme (SWEEPER GA No. 644313). The authors
would like to thank prof.dr. R. D. Howe and dr. D. Perrin for their input
of this research and making computing resources available. The authors
declare that tOpen asset ↗10.4121/uuid:884958f5-b868-46e1-b3d8-a0b5d91b02c0pdf-raw-page:12 lines:81-119Code / dataset availability confirmedEurope PMC · checked 15 Sept 2026
In viticulture, there are several applications where bud detection in vineyard images is a necessary task, susceptible of being automated through the use of computer vision methods. A common and effective family of visual detection algorithms are the scanning-window type, that slide a (usually) fixed size window along the original image, classifying each resulting windowed-patch as containing or not containing the target object. The simplicity of these algorithms finds its most challenging aspect in the classification stage. Interested in grapevine buds detection in natural field conditions, this paper presents a classification method for images of grapevine buds ranging 100–1600 pixels in diameter, captured in outdoor, under natural field conditions, in winter (i.e., no grape bunches, very few leaves, and dormant buds), without artificial background, and with minimum equipment requirements. The proposed method uses well-known computer vision technologies: Scale-Invariant Feature Transform for calculating low-level features, Bag of Features for building an image descriptor, and Support Vector Machines for training a classifier. When evaluated over images containing buds of at least 100 pixels in diameter, the approach achieves a recall higher than 0.9 and a precision of 0.86 over all windowed-patches covering the whole bud and down to 60% of it, and scaled up to window patches containing a proportion of 20–80% of bud versus background pixels. This robustness on the position and size of the window demonstrates its viability for use as the classification stage in a scanning-window detection algorithms.
Why it matches plant phenotyping methodsブドウ芽の画像検出を中心に、SIFT・Bag of Features・SVMによる植物器官の表現型取得手法を開発・評価しており、単なる生物学的実験での測定ではない。
abstractthis paper presents a classification method for images of grapevine buds
Reproduction assets foundThe paper's grapevine bud image datasets (labeled bud/non-bud patch corpus) and the authors' .Net image-manipulation/annotation software and code are publicly available at the authors' dharma.frm.utn.edu.ar URLs, as stated in footnotes and the discussion.Dataset · publicy the
268
region, with a pre-selected patch step size and dimensions. This method
269
works similarly to a scanning-window algorithm, but we limit it to scan in
270
a restricted region. With this procedure we could obtain a lot of examples,
271
orders of magnitude more than the bud patches.
272
3All images datasets available in http://dharma.frm.utn.edu.ar/papers/vise/bc/4.Net software and code available in http://dharma.frm.utn.edu.ar/papers/vise/bc/11Open asset ↗dharma.frm.utn.edu.arpdf-raw-page:11 lines:1-55Code · public269
works similarly to a scanning-window algorithm, but we limit it to scan in
270
a restricted region. With this procedure we could obtain a lot of examples,
271
orders of magnitude more than the bud patches.
272
3All images datasets available in http://dharma.frm.utn.edu.ar/papers/vise/bc/4.Net software and code available in http://dharma.frm.utn.edu.ar/papers/vise/bc/11Open asset ↗dharma.frm.utn.edu.arpdf-raw-page:11 lines:1-55