An improved YOLOv5s recognition and detection algorithm for floating objects on lake surface

To address the question of issue of misdetection and missed detection on the YOLOv5s algorithm for floating objects in lake surface scenarios, an improved lightweight YOLOv5s algorithm is proposed to improve detection of lake surface floating objects. In the Backbone layer, the improved lightweight ShufflenetV2_cssp network is adopted combined with the DSPPF_CS module. In the Neck layer, an improved RFBSD module is introduced, and meanwhile, the CIoU_SC loss function and the scale scaling mechanism are employed to optimize bounding box regression. Experimental results demonstrate that the improved lightweight YOLOv5s algorithm can effectively mitigate the false detection and missed detection issues in lake surface detection, especially can suppress the misjudgment of water surface ripples in strong reflection scenarios, and the missed detection rate is significantly reduced. While ensuring detection accuracy, the algorithm achieves the optimization of frame rate, providing technical support for the practical application of lake surface floating object detection.