Visualizing Explanations of DCNN-Based Plant Leaf Disease Detection Using Combined Grad-CAM and LRP

In agricultural sector, various Artificial Intelligence (AI) and Machine Learning (ML) techniques have been explored for plant disease detection. Despite the gain in the performance for plant leaf disease detection using Deep Convolutional Neural Networks (DCNNs), their interpretability for higher performance remains a challenge. Explainable AI (XAI) techniques, such as Gradient-weighted Class Activation Mapping (GradCAM) and Layer-wise Relevance Propagation (LRP), enhance model transparency but suffer from limitations in noise sensitivity, clarity, and robustness. In the proposed work, we have explored a novel approach that integrates GradCAM and LRP to improve visual explanations in plant disease classification. The method processes GradCAM outputs to reduce noise, applies element-wise multiplication with LRP-generated heatmaps, and enhances smoothness using Gaussian blur. Evaluations based on Robustness, Complexity, Faithfulness, Localization, and Randomization demonstrate that our approach outperforms standalone GradCAM and LRP, offering clearer and more reliable visualizations.