Real-Time Tea Leaf Disease Detection Using Deep Learning-Based Models

원문 초록

Tea leaf diseases pose a significant threat to crop productivity, highlighting the need for efficient and accurate detection methods. The lack of cost-effective, lightweight models for deployment on end devices limits real-time detection. This study addressed this by annotating and utilizing the previously unlabeled Tea Sickness Dataset for object detection and deploying fine-tuned models on mobile devices.The YOLO-NAS-s, YOLOv8n, YOLOv5nu, and SSD-MobileNetV2 models were fine-tuned using this dataset to detect diseased tea leaves, achieving state-of-the-art performance. Among them, YOLOv5nu achieved a maximum mAP@50 of 0.969 and an F1-score of 0.927, demonstrating exceptional accuracy. Its lightweight architecture ensures fast inference and low resource usage, offering a balance between performance and computational efficiency, making it well-suited for real-time deployment. After the models were evaluated, the two most lightweight models were deployed on mobile devices, demonstrating the feasibility of using high-performance and lightweight models for real-time plant disease monitoring.