Explainable AI (XAI) driven Mango leaf disease detection using State-ofthe-art convolutional neural network and Tiny-Net

Abstract

Rising agricultural requirements for intelligent, accessible solutions demonstrate why efficient plant disease detection systems have become vital. A deep learning framework has been developed to automatically detect mango leaf diseases while providing mobile access through an explainable framework solution. Six state-of-the-art convolutional neural network (CNN) models underwent initial evaluation on the MLD24 dataset, with DenseNet101 producing the most precise classification results of 99.06%. After researchers realised that traditional models were unsuitable for practical deployment due to their computational constraints, a custom lightweight model called Tiny-Net was developed. The Tiny-Net model delivered performance like standard models, achieving 99.61% accuracy while using minimal system resources, thus meeting the needs of mobile applications. The Explainable AI (XAI) methods Grad-CAM and LIME and SHAP provided transparent analysis of model decisions to increase system transparency and foster trust. The TensorFlow Lite version of the final model received implementation within a Flutter-based mobile application which delivered real-time offline detection capabilities accessible to farmers in regions with low connectivity. The system proved resistant to failure while demonstrating ease of use through multiple evaluations that proved its real-world utility. Through this research researchers demonstrated progress in automated plant disease detection technologies while using human-centered artificial intelligence deployments to benefit sustainable farming practices. The proposed system proves its ability to enhance farmers' disease diagnosing capabilities through real-time accurate diagnosis while closing the gap between controlled laboratory models and realistic usage conditions.