Lightweight traffic sign small object detection algorithm improved based on YOLOv8s

Traffic sign recognition technology has a fundamental role in the autonomous driving system, which is of major relevance for driving safety and security. However, because traffic signs are generally small-sized targets, they are prone to omission and misdetection, which lowers identification accuracy. In response to these challenges, an enhanced lightweight traffic sign recognition algorithm—YOLO-HRP—is built based on the YOLOv8s method. Combined with HetConv heterogeneous convolution, a new Het\_C2f module is designed to replace the existing C2f structure, the Het\_C2f module enables multi-scale and cross-receptive-field feature extraction using heterogeneous convolution, while simultaneously decreasing the computational cost and number of parameters of the model via lightweight operations. Based on the partial convolution (PConv) and CBAM attention mechanism, a novel detection head—PCB\_Head—is presented, which substantially boosts the model's capabilities for tiny target recognition and minimizes computational cost. By merging two attention processes, iRMB and EMA, a lightweight and efficient attention module dubbed RBMA is built, which enables the model to concurrently accomplish multiscale perception and local modeling, considerably enhancing its adaptation to complicated situations. Finally, by increasing the number of upsampling and prediction output layers, the model may collect more plentiful positional information, successfully resolving the issue of insufficient spatial detail extraction in the YOLOv8s model while handling tiny objects. Experiments are done using the Chinese traffic sign dataset TT100K. The experimental findings demonstrate that, compared to YOLOv8s, the proposed approach decreases the number of parameters by 31\%, the calculation by 29.5\%, and improves the recall (R) and mAP@50 by 1\% and 0.9\%, respectively. Compared with standard approaches, YOLO-HRP offers considerable performance benefits, enabling new opportunities to satisfy the demands of mobile deployment.