In urban traffic, safety accidents caused by electric bike riders often occur. Wearing a safety helmet can effectively avoid or reduce the damage caused by a safety accident. Therefore, many cities have promulgated relevant regulations on wearing a safety helmet. Aiming at the existing problem of low detection accuracy of helmet wearing, this paper proposes an algorithm for detecting the safety helmet wearing based on improved YOLOv3. The improved algorithm in this paper adopts the weighted feature fusion of channel and spatial attention modules, and combines densely connected networks to improve the effect of feature extraction, and adds a spatial pyramid pooling structure to enhance features. In this paper, the improved algorithm is tested and compared with original YOLOv3 at the self-built electric bike helmet wearing detection data set. The obtained results show that the mean average precision of the improved algorithm proposed in this paper reaches 93.29%, which is much higher than the original YOLOv3 algorithm. Experiments confirm that the proposed model can effectively enhance detection accuracy of electric bike helmet wearing detection.