A Highway Pavement Crack Identification Method Based on an Improved U-Net Model

Crack identification plays a vital role in preventive maintenance strategies during highway pavement maintenance. Therefore, accurate of cracks images is the key to work. In this paper, an improved U-Net network adopting multi-scale feature prediction fusion and parallel attention module was put forward better identify concrete cracks. Multiscale combines multiple features generated by intermediate layers for aggregated prediction, thus using global information from different scales. The used process decoded output fusion, which can give more weight target region image further capture contextual improve recognition accuracy. Improving bottleneck layer robustness model prevent overfitting. Experiments show that paper has significant improvement over original network. performance proposed method investigated on two publicly available datasets (Crack500 CFD) compared with competing methods literature. Using Crack500 dataset, achieved highest score precision (89.60%), recall (95.83%), mIOU (83.80%), F1-score (92.61%). Similarly, CFD high values (93.29%), (82.07%), (86.26%), (89.64%). Thus, several advantages identifying pavements ideal tool practical future work, crack types light-weighting are objectives. Meanwhile, provides new idea road identification.