Research on YOLOv5s Improved Algorithm for Pavement Crack Detection in Complex Environments

Abstract

Aiming to solve the problem of pavement crack detection in complex road environments, an improved algorithm based on YOLOv5s is proposed. First, the CBAM (Convolutional Block Attention Module) is introduced after the backbone network’s C3 modules to focus the complex scene’s practical information and enhance the model’s attention to the crack region. Second, the BiFPN (Bidirectional Feature Pyramid Network) is used in Neck to replace the bidirectional PANet (Path Aggregation Network) in YOLOv5s to improve the multiscale feature fusion, which reduces detection leakage due to illumination and scale factors. Thirdly, the P6 detection head for denser cracks is added to the Head, and the CA (Coordinate Attention) module is introduced to improve the crack detection capability at multiple scales. Finally, the algorithm is experimentally compared with SSD (Single Shot MultiBox Detector), Faster-RCNN (Faster-Region Convolutional Neural Network), and unimproved YOLOv5s on the constructed pavement crack image dataset. The results show that compared with other algorithms, when [email protected] (mean Average Precision when IoU=0.5) values are improved by 10.6%, 9.2%, and 4.6%, respectively, and cracks are identified better than the other three models.

Keywords

• Complex environments
• deep learning
• YOLOv5s
• pavement crack detection