The Explainable AI for mango leaf disease detection: bridging the gap between model accuracy and farmers usability

Abstract

Mango leaf diseases can seriously impact on the yield and vitality of mango trees, resulting in considerable financial losses. Prompt and precise identification of these diseases are essential for facilitating quick action and improving agricultural management practices. In the past few years, convolutional neural network (CNN) models have gained significant popularity towards image recognition and classification. Using CNN models, approaches for image-based disease diagnosis in the crops have become increasingly popular within the current scientific community. Mango leaves disease represents considerable threats to mango cultivation globally, making it essential to develop precise and efficient classification methods for timely disease control. Our research focuses on introducing an Explainable AI (XAI) framework that incorporates a modified VGG-16 CNN, alongside Gradient-weighted Class Activation Mapping (Grad-CAM), to recognize seven major mango leaf diseases using the publicly available MangoLeafBD dataset (3,500 images across seven classes). Our model demonstrated outstanding effectiveness in classification, achieving 92.8% accuracy, while as providing precise and graphical explanations to enhance use and foster farmer trust. Our results provide important insights for implementing CNN models that improve the accuracy and effectiveness of monitoring plant diseases in agricultural environments, ensuring greater clarity in model decision-making to optimize the framework for low-resource devices, expanding the dataset to include diverse mango varieties, and exploring multi-crop applications.