HPDD-YOLOv5: A Specialized YOLOv5 Network for Highway Pavement Defect Detection

Aiming at defect detection in highway pavement maintenance and management, a specialized network HPDD-YOLOv5 is proposed. Preselected anchors are designed to suit well for the specific task. CBAM is inserted into C3 of the Backbone to enable the network to focus on the useful information and enhance its ability of feature extraction. The ASFF is embedded in the Neck to make the feature maps of different levels fuse better and enhance the capability for feature fusion. An extra detection layer for more scale defect detection is added. Based on the RDD2022 data set, the mean average precision (mAP) of HPDD-YOLOv5 is 2.3% higher than that of YOLOv5, and the FPS reaches 68, which indicates that YOLOv5 has better detection accuracy and real-time performance.