A Dual-Net With Graph Attention Network for Plant Diseases Identification

Timely and accurate identification of plant diseases is essential for maximizing crop produce and supporting sustainable agricultural practices. In this study, we focus on classifying rice leaf diseases by applying a hybrid deep learning framework to distinguish between eight disease categories and healthy leaves. The proposed model combines a graph attention network (GAT) branch, which captures intricate relationships among image features, with a custom convolutional neural network (CNN) branch, which extracts critical spatial features. This dual-branch architecture leverages the strengths of both networks, utilizing relational and spatial information for enhanced classification accuracy. We employed a diverse dataset from online repositories, such as Kaggle and Mendeley, and augmented it to improve robustness. The performance of model is assessed on multiple disease types, confirming that the integration of spatial and graph-based feature extraction achieves significant improvements in disease detection accuracy. The proposed model achieved outstanding performance in classifying rice leaf diseases, with an accuracy of 99.28%, a precision of 98.79%, a recall of 98.81%, and an $F1$ -score of 98.80%. In addition to rice leaf disease, the model is evaluated on other plant types achieving an accuracy of 99.64% for apple, 99.25% for corn, and 99.59% for wheat. These results demonstrate the potential of hybrid deep learning solutions to provide reliable, scalable tools for automated plant diseases monitoring, ultimately aiding the agricultural sector.