Enhanced Deep Learning Framework for the Detection of Crop Leaf Diseases

Crop diseases significantly impact food production, making their early identification crucial for farmers. In India alone, over 43,000 acres are dedicated to stone fruit cultivation, producing approximately 0.25 million tons annually. Identifying diseases in fruit leaves with the naked eye poses challenges due to the complexity of damaged leaf images, which often exhibit irregular shapes, varying sizes, rich colors, blurry boundaries, and cluttered backgrounds. Effective disease management relies on precise segmentation of plant leaf images, a task that has become more feasible with advancements in deep learning and graphics processing unit (GPU) technology. This study proposes an enhanced ResUNet model aimed at improving the performance of deep convolutional neural networks while reducing the number of parameters. The model incorporates ResNet50’s skip connections in the encoder, enhancing the transfer of feature information between the contraction and expansion paths of the U-Net architecture. This customization improves the model's capability in segmenting diseased plant leaves. By combining convolutional blocks in the decoder with encoder skip connections, the model achieves superior expression ability. The proposed ResU-Net model demonstrated robust performance, achieving training, validation, and testing accuracies of 94.07%, 94.21%, and 95.00%, respectively. When compared to other existing methods, the results validate the model's effectiveness in addressing the plant leaf segmentation problem. This approach offers a promising solution for efficient and accurate identification of crop diseases, contributing to improved agricultural practices.