Download PDF
ABSTRACT
Aiming at the difficulty of rice multi-scale pest and disease identification and deployment of lightweight detection models in field environment, this paper proposes a lightweight rice pest and disease identification model SCR-YOLO based on the improved YOLOv11n. The model carries out a triple optimization on the basis of YOLOv11n, the RepViT module is introduced to enhance the feature representation capability by structural reparameterization technique. The CBAM hybrid attention mechanism is embedded to strengthen the attention to the key areas of spots; and the SimSPPF module is adopted to optimize the efficiency of multi-scale feature fusion. CBAM hybrid attention mechanism to strengthen the focus on the key areas of disease spots, and adopting SimSPPF module to optimize the efficiency of multi-scale feature fusion. The experimental results on four types of rice pests and diseases datasets containing leaf blight, rice blast, hoary mottle and rice fly show that SCR-YOLO achieves a significant lightweight effect while maintaining a high detection precision, with a precision rate (P) of 84.7%, a recall rate (R) of 84.2%, a mean average precision of 87.9% (mAP50), a reduction in the number of model parameters to 2.3 M, and the computational effort is only 7.3 GFLOPs.Deployment tests on a Jetson Nano embedded device show that the improved model's single-image inference time is significantly optimized, achieving a detection speed of 8.3 frames per second. This study provides an efficient and feasible lightweight solution for real-time accurate identification of rice pests and diseases, which is of positive significance for promoting the practical application of intelligent plant protection equipment in complex field environments.
Keywords:Rice Pest Identification; Target Detection; YOLO11; Attention Mechanism; Lightweight Structure