Research on traffic sign detection method based on FLB-YOLOv8

In this paper, the proposed FLB-YOLOv8 model address issues in current traffic sign recognition, such as leakage, false detection, low accuracy, and excessive model parameters. Firstly, a small target detection layer is added to improve semantic information and feature expression for small targets. Secondly, the FasterNet Block structure replaces the Bottleneck structure in the C2f module of YOLOv8, reducing parameters and computation while maintaining accuracy. The introduction of the LSK module adjusts the spatial receptive field dynamically, improving feature extraction capability. Lastly, a weighted bidirectional feature pyramid network enables efficient bidirectional cross-scale connection and feature fusion, balancing feature information of different scales. Experimental results on the CCTSDB dataset show that the improved model achieves 78.3% mean average precision (mAP), a 4.7 percentage point improvement over the original model. It also reduces model parameters by 56% and computation by 26%. This algorithm effectively improves detection accuracy, reduces model complexity, and achieves better performance in traffic sign detection.