Enhanced YOLO Optimization Study for Road Sign Detection in Autonomous Driving

This study aims to propose an improved YOLO series algorithm named YOLO-PAN, which is used to assist automatic driving in road sign detection, and its performance is superior to the basic YOLOv8 algorithm in all aspects. In this study, the YOLO-PAN model is improved on the basic YOLOv8 model and introduces the LSKA module and SPPELAN module to replace the original SPPF module. The LSKA megacore attention mechanism is designed to improve the model’s localization and scale perception ability of relevant small targets in the scene, thereby enhancing the accuracy and stability of target detection. At the same time, the SPPELAN module further optimizes the performance of object detection by introducing an improved spatial pyramid pool structure. The road sign dataset on the Kaggle platform was used for experimental evaluation and compared with the basic YOLOv8 model and the improved model of related YOLOv8. The experimental results show that the YOLO-PAN model has achieved remarkable performance improvement in the road sign detection task, and the accuracy rate increases by more than $6 \%$, which proves the effectiveness and feasibility of the proposed method.