Addressing the challenges of sluggish detection speed and suboptimal detection accuracy, this paper introduce the improved YOLOv8s, an innovative rice disease detection algorithm that builds upon and refines the YOLOv8s framework. This algorithm incorporates several strategic enhancements to bolster its performance. To bolster the network capacity in identifying the target, the CBAM attention mechanism is first into the final layer of YOLOv8s feature extraction network. Additionally, to further refine the model ability to generalize, BiFPN feature fusion network was introduced to achieve effective multi-scale feature fusion and balance of efficient computing performance, and improve the algorithm proficiency to distinguish leaf diseases. Finally, WIoU v3 loss function is introduced to reduce the harmful gradient ino extreme samples. After the comparison experiment of the model, the improved algorithm (CBW-YOLOv8s) showed better performance in rice disease detection. In comparison to the original algorithm, the enhanced version reaches mAP value of 90.2%, which is an increase of 3.7%. The improved model has better performance than other common algorithms, and can provide reference for rice disease detection in complex field environment.