A ResNet50-DPA model for tomato leaf disease identification.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Tomato leaf disease identification is challenging due to disease variety and complexity.
• Convolutional neural networks (CNNs) show promise but struggle with feature extraction, leading to low accuracy.
• Existing methods often lose critical features during image analysis.

Purpose of the Study:

• To propose an improved ResNet50 model integrated with a dual-path attention (DPA) mechanism for accurate tomato leaf disease identification.
• To enhance feature extraction capabilities for better disease detection.
• To improve the interpretability and accuracy of automated plant disease diagnosis systems.

Main Methods:

• An improved ResNet50 architecture incorporating cascaded atrous convolution in the initial layer to capture multi-scale leaf features.
• A novel dual-path attention (DPA) mechanism utilizing stochastic pooling and 1D convolutions to identify key features and minimize information loss.
• Integration of the DPA module within the ResNet50 residual blocks to generate enhanced feature maps.

Main Results:

• The proposed ResNet50-DPA model demonstrated higher accuracy in identifying tomato leaf diseases compared to standard methods.
• The DPA mechanism effectively captured salient features, reducing the loss of critical information.
• Grad-CAM visualization confirmed the model's improved interpretability and accurate disease localization.

Conclusions:

• The ResNet50-DPA model offers a robust and accurate solution for tomato leaf disease identification.
• The integration of atrous convolution and DPA mechanism significantly enhances feature extraction and diagnostic accuracy.
• This approach holds potential for reducing economic losses in agriculture through precise disease management.