Automatic Detection of Flavescence Dorée Symptoms Across White Grapevine Varieties Using Deep Learning.

Area of Science:

• Plant Pathology
• Agricultural Technology
• Computer Science

Background:

• Flavescence dorée (FD) is a destructive grapevine disease spread by leafhoppers in Europe.
• Current control methods are challenged by the disease's spread and complex symptom expression.
• Early and accurate detection is crucial to prevent further contamination and economic loss.

Purpose of the Study:

• To develop an automated detection model for Flavescence dorée-like symptoms in vineyards.
• To enable rapid identification and removal of infected grapevines, preventing disease spread.
• To leverage deep learning for efficient and accurate symptom detection under field conditions.

Main Methods:

• Utilized deep learning, specifically training a Convolutional Neural Network (CNN) on image patches.
• Converted the CNN into a Fully Convolutional Network (FCN) for coarse symptom area segmentation.
• Evaluated model performance across multiple grape varieties, including Chardonnay and Ugni-Blanc.
• Employed visualization techniques (Grad-CAM, UMAP) for model transparency and sensitivity analysis.

Main Results:

• The model achieved a high true positive rate (98.48%) on Chardonnay but significantly lower on Ugni-Blanc (8.3%).
• Results highlight the critical need for multi-varietal training datasets to account for diverse symptom expressions.
• Visualization techniques provided insights into model behavior and reliability for in-field applications.

Conclusions:

• Deep learning offers a powerful approach for automated detection of grapevine yellows symptoms.
• Model performance is highly dependent on the diversity of the training data, necessitating multi-varietal datasets.
• Further development with diverse datasets is essential for reliable, widespread application in vineyard disease management.