Plant Disease Detection using Advanced Transfer Learning and Attention Module

Plant diseases significantly threaten global food security, reducing agricultural productivity and causing economic losses for farmers. Early detection and effective disease management are crucial to mitigating these impacts. In this study, we introduce an advanced deep learning model, InceptionResNetV2, enhanced by the Convolutional Block Attention Module (CBAM). This technique is aimed at enhancing feature extraction for the classification of plant diseases. The model leverages spatial and channel-wise attention mechanisms to enhance detection accuracy. In addition, Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations provide interpretability by highlighting key regions that influence predictions. We conducted a comparative analysis against other deep learning models, including a customdesigned Convolutional Neural Network (CNN), InceptionResNetV2, and VGG16 using a multi-class agricultural disease dataset. Extensive data augmentation techniques were applied to improve model generalization. Experimentation demonstrates that the proposed CBAM-augmented InceptionResNetV2 model outperforms other approaches, achieving a validation accuracy of 98.23%, making it a promising scheme for plant disease identification.