Research on a Method for Identification of Chinese Rose Leaf Pests and Diseases Based on a Lightweight CR-YOLO Model

Accurate and rapid detection of pests and diseases on Chinese rose leaves is crucial for horticultural management and production quality. Despite advances in detection methods, challenges such as complex backgrounds, variable lighting conditions, and subtle disease manifestations in natural environments often lead to diminished detection accuracy and high computational costs. Traditional detection models typically require substantial computational resources, limiting their practical applicability in real-world horticultural settings. In response to these challenges, we introduce CR-YOLO (Chinese Rose YOLOv7-tiny), an enhanced lightweight detection framework derived from YOLOv7-tiny. Replacement of standard convolutions with distributed offset convolutions in the backbone network reduces computational complexity while preserving feature extraction capabilities, integrating SimAM (parameter-free attention) to boost feature discrimination, and adopting Focal-EIOU loss to accelerate convergence. Experimental results on a comprehensive Chinese rose leaf pest dataset collected under natural conditions demonstrate that CR-YOLO achieves impressive metrics with precision, recall, mAP_0.5, and mAP_0.5-0.95 of 95.1, 94.5, 97.6, and 93.7%, respectively. Compared with YOLOv7-tiny, CR-YOLO improves key metrics by 11.8 to 26.7 percentage points. When evaluated against other prevalent detection models, CR-YOLO shows superior performance with mAP_0.5 improvements of 6.9 to 11.4 percentage points while reducing computational overhead by 3.8 to 31.5%. The model's exceptional performance and lightweight architecture make it particularly suitable for resource-constrained platforms, enabling real-time monitoring in practical horticultural applications and advancing precision agriculture through efficient, automated pest and disease detection.