YOLO-LeafNet: A Robust Deep Learning Framework for Real-Time Plant Disease Detection
Abstract
Plant diseases are a risk to agricultural productivity and require accurate and real-time detection of the disease under various field conditions. In this study, a novel YOLO-based plant disease detection framework which uses attention mechanism is proposed called YOLO-LeafNet, focusing on the multi-crop plant disease detection, especially on underexplored crop Tea. A novel framework incorporates image preprocessing techniques such as auto-orientation, brightness normalization, and data augmentation to enhance feature representation and lesion localization through the use of a Convolutional Block Attention Module (CBAM). This model was trained on 12,000 images of different crop classes with annotations, and tested in challenging environmental conditions. YOLO-LeafNet was able to achieve an mAP@0.5 of 0.945, F1-score of 0.952 and real-time inference of about 42 FPS, which are better than the baseline and lightweight YOLO models. Evaluation and robustness analyses over cross datasets with occlusion, low illumination and low resolution conditions show strong generalization and deployment ability in precision agriculture applications.