Discovering the Depths of Cotton Leaf Disease Detection: Integrating Hypertuned Residual Networks with GradCAM XAI for Enhanced Understanding and Diagnosis

Cotton plays a pivotal role in the textile, medical, and oil industries and is considered the world's leading cash crop. Numerous agriculture-based countries, textile sectors, and people's livelihoods rely heavily on cotton for their economic importance. However, the hazards posed by pests, fungi, viruses, and bacteria forming different cotton leaf diseases enormously reduce cotton production. Hence, prompt detection and identification using advanced machine learning technology has assumed enormous importance for detecting crop diseases for agricultural sustainability and the broader economic landscape. Therefore, this paper is going to present a perceptive comparative analysis of Residual Neural Networks (ResNets) with Grad-CAM XAI (Explainable AI) heatmaps to identify the most effective model used to address the pressing issue of cotton leaf disease detection. Specifically, we compare and evaluate six sophisticated ResNet transfer learning models to foster a deeper understanding of their strengths and limitations, paving the way for advancements in cost-effective, efficient, and accurate disease detection. We demonstrate a clear understanding that ResNet101V2 has the best performance (96.9%) for its model depth, skip connections, and other architectural nuances when comparing other ResNet models. Also, we integrate the Grad-CAM XAI model fostering transparency, understandability, and insights into the decision-making processes of models for facilitating collaboration with humans. By undertaking this comparative and explainable investigation, we aim to furnish a comprehensive and optimal model for the accurate identification in agricultural management practices particularly for cotton leaves disease detection to avoid the lucky choice.