Abstract: To address the road damage detection problem, an improved YOLOv5-based asphalt pavement crack detection method was proposed to overcome the disadvantages of high cost and low efficiency of manual asphalt pavement crack detection. Firstly, asphalt pavement images were collected by a multifunctional inspection tool, and 5 224 images were divided into training and test sets at a 7:3 ratio to construct the asphalt pavement crack detection dataset. Then, the CBAM module, BiFPN module, and GSCConv network were introduced to modify the feature extraction network of the YOLOv5 model for asphalt pavement crack detection. The experimental results show that the F1 value, accuracy, mAP, and recall of the improved YOLOv5 algorithm model increase by 0.4%, 2.4%, 1.2%, and 1% respectively, and the number of model parameters decreased by 1.1×10⁶ compared with the original algorithm model, and a detection accuracy of 88.3% is obtained for the improved model. Therefore, the improved YOLOv5 model can effectively detect various types cracks in complex road scenes.