Pavement damage detection model based on improved YOLOv5

In order to further improve the detection accuracy of pavement damage, the detection model named Pavement Damage⁃YOLO (PD⁃YOLO) based on YOLOv5 for pavement damage characteristics is proposed in this paper. PD⁃YOLO introduces the Space⁃to⁃ depth layer into the network structure to adapt to the detection tasks with low resolution and small pavement damage targets. In addition, at the pool level, SPPFCSPC algorithm is used in PD⁃YOLO to obtain different receptive fields in feature extraction, which effectively solves the problem of large difference of target size in pavement damage detection images. At the feature fusion level, ASFF module is introduced to make the model learn the relationship between different features adaptively, so as to strengthen the attention of the model to the damage target area. In multiple sets of test datasets, compared with YOLOv5, the PD⁃YOLO model can improve the accuracy, recall rate, F1 value and mAP @ 0. 5 value of the test results. The research results show that PD⁃YOLO has stronger feature extraction ability and stronger feature fusion ability, and has better performance in pavement damage detection. © 2024 Chinese Society of Civil Engineering. All rights reserved.