GNN-based Disease Detection and Classification in Cassava Leaf

Cassava, a staple crop for millions of people worldwide, is highly susceptible to various leaf diseases, which can significantly reduce crop yields. Detecting and classifying these diseases at an early stage is crucial for effective disease management and crop protection. In this paper, we propose a novel approach for cassava leaf disease classification by combining Convolutional Neural Networks (CNNs) and Graph Neural Networks (GNNs). While CNNs excel at extracting local features from images, they often fail to capture complex spatial relationships between different regions of the leaf. To address this limitation, we integrate GNNs, which are well-suited for learning from irregular, structured data, such as the vein patterns and disease spread structures found in cassava leaves. By modeling the relationship between different regions of the cassava leaf as a graph G (V, E), where V represents nodes corresponding to superpixel regions of the leaf, and E represents edges capturing spatial relationships, the hybrid CNN-GNN model offers a more comprehensive and accurate classification system. The model is trained and evaluated on a dataset of cassava leaf images covering multiple disease categories, demonstrating superior performance over traditional CNN-based image classification techniques. This approach provides a scalable solution for improving the detection and management of cassava leaf diseases.