Multi-Class Pavement Disease Recognition Using Object Detection and Segmentation

Abstract

Pavement disease is an important factor threatening road safety. Most traditional disease recognition methods often rely on manual detection, which is time-consuming and inefficient. In this work, by introducing the object detection and segmentation into the detection of pavement diseases, a multi-class pavement disease detection method is proposed. First, diseases are located based on YOLOv4. CSPDarknet53 is used as the backbone network. The feature extraction performance is further improved by spatial pyramid pooling. Then, on the basis of pavement disease location, the pyramid scene parsing network (PSPNet) is employed to extract the pixel of the disease area to realize the accurate analysis of the anomaly. The feasibility of the proposed method is verified by a pavement disease detection experiment using the actual road dataset collected from a province in eastern China, including seven common diseases.