A robust deep learning approach for tomato plant leaf disease localization and classification.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Early detection of tomato plant diseases is crucial for preventing crop loss and increasing food production.
• Classifying tomato leaf diseases is challenging due to visual similarities between healthy and diseased areas and low contrast.
• Existing methods struggle with timely localization and identification of diverse tomato leaf diseases.

Purpose of the Study:

• To develop a robust deep learning approach for accurate tomato plant leaf disease classification.
• To address challenges in disease localization and identification caused by image variations and low contrast.
• To provide an automated, cost-effective solution for farmers to manage tomato crop health.

Main Methods:

• A deep learning (DL) model, Faster R-CNN, was employed for tomato plant leaf disease classification.
• ResNet-34 integrated with the Convolutional Block Attention Module (CBAM) served as the feature extractor.
• The process involved image annotation for region of interest (RoI) identification, feature extraction, and model training.

Main Results:

• The proposed ResNet-34-based Faster-RCNN achieved high performance on the PlantVillage Kaggle dataset.
• Achieved a mean Average Precision (mAP) of 0.981 and an accuracy of 99.97% with a test time of 0.23 seconds.
• Demonstrated robustness against variations in size, color, orientation, blurring, noise, and lighting conditions.

Conclusions:

• The developed deep learning framework offers a robust and accurate solution for tomato leaf disease detection and classification.
• The method significantly outperforms manual inspection and can be a cost-effective alternative for farmers.
• Future work aims to extend the approach to identify diseases on other plant parts.