Research on Real-time Detection Method of Rice Diseases and Pests Based on Improved YOLOv10s Model

As one of the world's most important food crops, rice's yield and quality are largely threatened by pests and diseases, necessitating a highly accurate, fast-responding, and adaptable intelligent detection method to support pest and disease control and sustainable agricultural development. This paper addresses the problems of low accuracy, poor real-time performance, and sensitivity to occlusion interference in complex field environments with existing methods, proposing a real-time detection method for rice pests and diseases based on an improved YOLOv10s model. First, a high-quality image dataset of 3100 images covering seven typical pests and diseases—rice blast, bacterial leaf blight, sheath blight, leaf spot, rice stem borer, rice borer moth, and rice green bug—was constructed, encompassing various real-world scenarios. A stratified sampling combined with five-fold cross-validation strategy was used for data partitioning to enhance the model's generalization ability. Secondly, to address the limitations of the receptive field and feature modeling in the YOLOv10s model, SAConv and SEAM structures were introduced. The former enhances spatial modeling capabilities through Multi-holes Rate Switchable Convolution Module, while the latter enhances the perception of fine-grained features and occluded targets based on a channel-Spatial joint attention mechanism. Furthermore, the MPDIoU bounding box regression loss function was introduced to optimize target box position prediction from a geometric perspective, significantly improving localization accuracy and training stability. Experimental results show that the improved model achieves an mAP of 93.1% and an inference speed of 176.4 FPS while maintaining a relatively low computational cost (20.3 GFLOPs) and parameter size (7.1 MB), significantly outperforming mainstream detection models such as YOLOv5s, YOLOv8s, and Faster-RCNN. It possesses good engineering deployment potential and field application value, providing an efficient and scalable solution for intelligent identification of agricultural pests and diseases.