LeafXNet: A Lightweight Explainable Model With Integrated Attention Mechanism for Fruit Leaf Disease Detection

Food security is threatened by climate change, altered precipitation patterns, soil degradation, and overuse of pesticides. These challenges compromise the sustainability of agricultural practices and increase the financial burden on farmers. Fruits such as watermelons, grapes, and guava, which are rich in nutrients, play a crucial role in global diets and economies. However, their production often faces challenges owing to emerging diseases and inadequate detection techniques. Conventional approaches to fruit leaf disease management remain inadequate, particularly in settings with limited resources. Consequently, there is a pressing need for reliable, economical, and scalable detection systems to safeguard agricultural outputs and improve global food security. To address these challenges, we introduced LeafXNet, a lightweight attention-enhanced convolutional neural network (CNN) with only 79,717 parameters for fruit leaf disease detection. It demonstrates effective generalization and resource-efficient performance suitable for web, mobile, and Internet of Things applications. We integrate three distinct methods of Explainable AI (XAI) to enhance understanding and foster trust in the predictions made by the model. The results demonstrate that LeafXNet surpasses state-of-the-art pretrained CNN architectures while utilizing substantially fewer computational resources, achieving accuracies of 99.74 ± 0.0032, 99.76 ± 0.0029, and 97.59 ± 0.0086 for watermelon, grape, and guava leaf disease classification, respectively. However, the model is limited to disease-versus-healthy classification and requires validation on visually similar nondisease conditions before real-world deployment.