A FEATURE-ENHANCED EFFICIENTNETV2 FRAMEWORK FOR INTELLIGENT DETECTION AND CLASSIFICATION OF COFFEE LEAF DISEASES IN PRECISION AGRICULTURE

Abstract

Coffee is a globally significant cash crop and a key driver of India’s agricultural economy. However, its yield and quality are substantially affected by a range of foliar diseases such as Hemileia vastatrix (rust), Leucoptera coffeella (leaf miner), and Phoma costarricensis, which often remain undetected until advanced infection stages. To address this challenge, this study introduces a segmentation-guided deep learning architecture that combines the U²-Net segmentation framework with an enhanced EfficientNetV2 classifier for precise detection and categorization of coffee leaf diseases. The proposed pipeline employs U²-Net to isolate diseased regions, suppress background interference, and enhance the discriminative features of infected leaf areas before classification. Subsequently, an Intensified EfficientNetV2 model, fine-tuned through transfer learning, is utilized to extract multi-scale spatial features with improved convergence stability and reduced computational overhead. The model achieves 100% classification accuracy during progressive training, outperforming the DenseNet-121 baseline (98.9%) while using fewer parameters, thereby demonstrating both computational efficiency and superior generalization. The system effectively distinguishes four disease classes and healthy samples, enabling real-time deployment via a web-based interface for field-level disease surveillance. The proposed framework offers a scalable and interpretable solution for precision agriculture, supporting sustainable disease management and yield optimization in coffee plantations.