Asphalt Pavement Crack Detection Based on Improved YOLOv5s Algorithm

Taking asphalt pavement crack as the research object, the aim is to find pavement distress in time and make targeted maintenance decisions. The method is based on YOLOv5s neural network structure, and provides reference for asphalt pavement distress recognition. Adaptive attention and feature enhancement module were added in the process of feature extraction, which can effectively improve the accuracy of crack identification by reducing the feature channel and enhancing the feature pyramid. At the same time, the accuracy of crack detection and location can be effectively improved by reconstructing the feature pyramid structure to solve the challenge of crack sizes' difference. Experiments show that the improved algorithm can identify the crack with an accuracy of 90.75%, which is 11.58%, 8.9% and 1.98% higher than YOLOv3, YOLOv4 and YOLOv5s, respectively. The results show that the improved YOLOv5s algorithm has the better accuracy in asphalt pavement crack detection.