LEAF: A Lightweight Explainable AI Framework for Mango Anthracnose Detection and Severity Grading

Mango anthracnose, caused by Colletotrichum gloeosporioides, is a destructive fungal disease that severely reduces fruit quality and yield, threatening farmer livelihoods and the mango industry. Early and reliable detection is crucial, as traditional manual inspection is slow, subjective, and often fails to identify infections at an early stage. To address this challenge, we present LEAF: a Lightweight Explainable AI Framework for mango anthracnose detection and severity estimation. LEAF employs a fine-tuned MobileNetV2 model for binary classification of healthy and infected mango leaves, coupled with severity grading to provide actionable insights. The framework integrates Grad-CAM for explainability, allowing visualization of infection regions and improving model interpretability. Experimental evaluation on a curated dataset achieved over 99% classification accuracy using MobileNetV2 with transfer learning. The lightweight architecture, further optimized through TensorFlow Lite, demonstrates suitability for real-time mobile deployment, while Grad-CAM enhances interpretability by localizing disease-specific regions in mango leaves. The combined use of efficient transfer learning, explainable AI, and severity quantification highlights the novelty of this approach and its practical value in field applications.