PCSD-YOLOv11s: Target Detection Algorithm in Traffic Scene for Complex Weather Conditions

To improve the problems of missed and false detection of targets and insufficient fusion of target features in traffic scenes with complex and severe weather conditions in most existing computer vision algorithms, this paper proposes the PCSD-YOLOv11s object detection algorithm based on the YOLOv11 network. Firstly, PConv convolution is used to extract multi-directional target features. Secondly, build F-C3k2 module based on C3k2, significantly reduce the number of model parameters, save computing resources, and improve computing speed. Furthermore, the P2 detection layer is introduced and combined with the SSFF module to enhance the network's ability to fuse features of different scales. Finally, a lightweight DSCD detection head is designed to enhance the detection accuracy of small and occluded targets. The experimental results on the BDD100K dataset show that compared with the original model, PSCD-YOLOv11s reduces parameters by 1.111 M and reduces model size by 0.5 MB, improved mAP50 by 6.6 %, it can effectively complete target detection in traffic scenes under complex weather conditions, improving the algorithm's generalization and robustness.