In order to improve the real-time and feasibility of traffic sign detection for autonomous driving in complex traffic environments, this paper proposes a small target detection algorithm for traffic signs based on the YOLOv8 model. First, the bottleneck of the C2f module in the original yolov8 network is replaced with the residual Faster-Block module in FasterNet, and then the new channel mixer convolution GLU (CGLU) in TransNeXt is combined with it to construct the C2f-faster-CGLU module, reducing the number of model parameters and computational load
Secondly, the SPPF module is combined with the large separable kernel attention (LSKA) to construct the SPPF-LSKA module, which greatly enhances the feature extraction ability of the model
Then, by adding a small target detection layer, the accuracy of small target detection such as traffic signs is greatly improved
Finally, the Inner-IoU and MPDIoU loss functions are integrated to construct WISE-Inner-MPDIoU, which replaces the original CIoU loss function, thereby improving the calculation accuracy. The model has been validated on two datasets Tsinghua-Tencent 100K (TT100K) and CSUST Chinese Traffic Sign Detection Benchmark 2021 (CCTSDB 2021), achieving Map50 of 89.8% and 98.9% respectively. The model achieves precision on par with existing mainstream algorithms, while being simpler, significantly reducing computational requirements, and being more suitable for small target detection tasks. The source code and test results of the models used in this study are available at https://github.com/lyzzzzyy/CSW-YOLO.git.