Enhancing leaf disease classification using GAT-GCN hybrid model.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate and timely plant disease detection is crucial for global food security and agricultural productivity.
• Existing methods often lack the precision and efficiency required for large-scale crop monitoring.
• The need for advanced computational models to analyze plant health is increasing.

Purpose of the Study:

• To develop and evaluate a hybrid Graph Attention Network (GAT) and Graph Convolution Network (GCN) model for accurate leaf disease classification.
• To enhance feature extraction and model generalization for robust plant disease identification.
• To assess the performance of the hybrid model against individual GCN and GAT models.

Main Methods:

• Utilized a hybrid model integrating Graph Attention Network (GAT) and Graph Convolution Network (GCN) for leaf disease classification.
• Employed superpixel segmentation for efficient feature extraction from plant leaf images.
• Incorporated edge augmentation techniques and weight initialization for improved model robustness and generalization.
• Evaluated the model on datasets of apple, potato, and sugarcane leaves.

Main Results:

• The hybrid GAT-GCN model achieved high performance metrics across different plant species.
• Achieved precision, recall, and F1-scores above 0.97 for apple and potato leaf disease classification.
• Demonstrated strong performance with precision, recall, and F1-scores of approximately 0.88 for sugarcane leaf disease classification.
• Outperformed individual GCN and GAT models in accuracy and consistency.

Conclusions:

• The hybrid GAT-GCN model offers a highly effective and consistent solution for plant leaf disease identification.
• The integration of GAT and GCN, coupled with advanced image processing techniques, significantly boosts classification accuracy.
• This research contributes a valuable tool for precision agriculture, aiding in crop monitoring and disease management.