Using a Resnet50 with a Kernel Attention Mechanism for Rice Disease Diagnosis.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Plant diseases cause significant global economic losses, with rice crops losing 20-40% of production.
• Accurate and timely diagnosis of rice diseases is crucial for effective management and mitigating financial impacts.
• Current rice disease diagnosis relies heavily on manual methods, despite technological progress.

Purpose of the Study:

• To develop and evaluate a novel AI-assisted system for accurate rice disease classification.
• To enhance the feature extraction capabilities of deep learning models for plant pathology.
• To improve the efficiency and effectiveness of agricultural disease diagnosis and management.

Main Methods:

• A self-attention network (SANET) was developed, integrating a kernel attention mechanism with the ResNet50 architecture.
• Attention modules were utilized to capture contextual dependencies and focus on critical features within rice leaf images.
• Cross-validated classification experiments were performed on a publicly available rice disease dataset.

Main Results:

• The attention-based mechanism effectively guided the convolutional neural network (CNN) in learning discriminative features.
• The proposed SANET model achieved a high test set accuracy of 98.71% for rice disease classification.
• The model demonstrated reduced performance variation and outperformed existing state-of-the-art methods.

Conclusions:

• The developed SANET model offers a highly accurate and reliable solution for AI-assisted rice disease diagnosis.
• The findings underscore the potential of AI, particularly attention mechanisms, to revolutionize agricultural disease management.
• Widespread adoption of such AI tools can significantly enhance efficiency and effectiveness in the agricultural sector.