A improved YOLOv5 asphalt pavement crack detection method is proposed to address the i-ssues of complex crack image backgrounds, small detection targets, poor detection performance, and missed detections in YOLOv5 crack detection. Firstly, the lightweight Mobilenet v3 network, as the feature extraction network of YOLOv5, is used to reduce the complexity of the model and speed up reaso-ning. Secondly, an efficient channel attention mechanism (CBAM) is employed to enhance the network’s ability to capture and fuse local features. Finally, an embedded Panet module is used to enhance the multi-scale feature expression ability of crack images and improve the detection performance of small targets. The experimental results show that compared to the original YOLOv5 algorithm, the improved YOLOv5 algorithm improves the mAP of asphalt pavement crack detection by 5.5%, reduces the number of model parameters by 86.3%, and reduces image detection time by 75.8%.