Traffic Sign Detection Based on Improved Yolov7

Traffic sign detection in the natural environment with small targets, weak light and other influencing factors often make the neural network has low accuracy in the detection process. In view of the above problems, this paper proposes an improved YOLOv7-WCN network based on YOLOv7. Firstly, the improvement of the YOLOv7 network is added to the backbone network to enrich the image information In terms of improvements to the YOLO v7 network, CHB module is added to the backbone network in order to enrich the image information. Then, the normalization-based attention module (NAM) is introduced to make the network pay more attention to the target region for the purpose of enhancement. Moreover, the CIOU loss function of YOLOv7 is replaced with the Wasserstein Distance Loss to optimize the algorithm training process. Experimental results show that the detection accuracy of YOLOv7-WCN improves from 85.5% to 89.0% compared with the original algorithm, which is 3.5 percentage points better.