Research on YOLOv8 Traffic Sign Detection with Convolutional Block Attention Module

Traffic sign detection is a critical component in autonomous driving environmental perception. However, challenges such as small targets and background interference in complex road scenes pose significant obstacles to the accuracy of detection. To address this, this paper proposes an improved YOLOv8 traffic sign detection algorithm. The algorithm embeds lightweight convolutional block attention modules after the P3 and P4 feature layers—used for small and medium object detection, respectively—in the feature pyramid of the YOLOv8 backbone network and neck network. This constructs the YOLOv8-CBAM model. This design aims to guide the network to adaptively enhance channel responses and spatial focus on key features of traffic signs, thereby improving the model's feature extraction capabilities. Experimental results on the TT100K dataset demonstrate comprehensive improvements in the modified model's precision, recall, and mAP@0.5和mAP@0.5:0.95 metrics. This effectively validates the CBAM module's efficacy in optimizing feature representation and enhancing traffic sign detection performance in complex scenarios.