DeepLeafDiagnosis: A Two-Phase Transfer Learning Strategy for Perfect Accuracy in Rice Leaf Disease Classification

Rice, a staple food for over half the global population, faces significant yield losses due to various leaf diseases. Early and accurate detection of these diseases is crucial for effective management and sustainable crop production. This study presents a novel approach to rice leaf disease classification using deep learning techniques. We develop and evaluate two deep convolutional neural network models based on VGG16 and ResNet50 architectures, employing a two-phase training strategy that combines transfer learning with systematic fine-tuning. Our methodology utilizes a comprehensive dataset of 5,932 images representing four major rice leaf diseases: Bacterial Blight, Blast, Brown Spot, and Tungro. The proposed models incorporate extensive data augmentation, dropout regularization, and batch normalization to enhance generalization capabilities. Performance evaluation reveals exceptional classification accuracy, with the fine-tuned VGG16 model achieving 100% accuracy and the ResNet50 model reaching 99.97% accuracy on the test dataset. Detailed analysis through confusion matrices, classification reports, and ROC curves confirms balanced performance across all disease categories. The proposed approach demonstrates the significant potential of optimized deep learning architectures for practical deployment in agricultural disease diagnosis systems, potentially transforming early disease detection capabilities in rice cultivation. This research contributes to the advancement of smart agriculture by providing a robust technological solution for automated rice disease monitoring that could significantly enhance crop management practices and food security.