Hybrid Deep Learning Architecture with Attention Mechanisms for Enhanced Plant Leaf Disease Detection and Classification

Plant leaf diseases pose a major challenge in agriculture, requiring early detection to prevent losses and ensure food security. In this study, a hybrid deep learning model is proposed to improve the classification of 35 plant leaf disease categories. The framework integrates two pre-trained Convolutional Neural Network (CNN) models—DenseNet121 and MobileNetV2—along with a custom model, all enhanced by attention mechanisms for better feature extraction. Fine-tuning was performed using attention modules and global average pooling layers, followed by fully connected layers with Swish activation, dropout, batch normalization, and L2 regularization. The model was trained on the New Plant Diseases Dataset from Kaggle and achieved higher accuracy, precision, recall, and Fl-score compared to baseline models. Grad-CAM (Gradient-weighted Class Activation Mapping) was used to identify the most influential regions in the leaf images, improving the model's explainability and reliability. This hybrid approach shows promise for practical application in agricultural disease detection systems.