Research on Urban Road Pavement Distress Detection System Based on Improved YOLOv4

According to the 14th Five-Year Plan in China, which proposes to accelerate the development of intelligent urban construction and service systems, urban roads are a vital component of smart city development. Road damage detection is an essential approach to ensuring effective management of city roads. In this paper, an improved YOLOv4 algorithm is employed to recognize road damage conditions through processes such as dataset creation, experimental setup, model training, and problem identification. Upon identification, the system promptly notifies maintenance personnel of necessary repairs, significantly reducing manpower and time costs. The dataset consisting of over 3500 urban road images was annotated, classified, and used for model training. Experimental results indicate that the proposed model achieves an average Mean Average Precision (MAP) of over 95%. Moreover, the model training speed is increased by 40%. The proposed approach can significantly improve the detection accuracy and speed of urban road defects.