Classification of Rice Leaf Diseases Using Deep Residual Graph Neural Networks

Decreased quality and quantity in agricultural production are often attributed to leaf diseases in rice plants. With the continuous evolution in the structure and cultivation methods of rice plants, there’s a surge in new leaf diseases. Identifying these diseases accurately at early stages is crucial to prevent their spread and to promote the growth of healthy plants. In this context, we introduce an innovative lightweight residual deep graph convolutional network (Res-GCN) model, designed to extract complex hidden features. This model blends deep feature learning with classical handcrafted feature patterns, enhancing the detection of local textures in images of rice plant leaves. The model has undergone training and testing on datasets that are available to the public. It has shown remarkable results, achieving 98.32% accuracy in validation and 98.8% in testing. These results suggest that our model presents a more effective strategy for managing diseases in plants.