Deep Learning-Based Semantic Segmentation Methods for Pavement Cracks

As road mileage continues to expand, the number of disasters caused by expanding pavement cracks is increasing. Two main methods, image processing and deep learning, are used detect these improve efficiency quality crack segmentation. The classical segmentation network, UNet, has a poor ability extract target edge information small segmentation, susceptible influence distracting objects in environment, thus failing better segment tiny on pavement. To resolve this problem, we propose U-shaped ALP-UNet, which adds an attention module each encoding layer. In decoding phase, incorporated Laplacian pyramid make feature map contain more boundary information. We also adding PAN auxiliary head provide additional loss for backbone overall network effect. experimental results show that proposed method can effectively reduce interference other factors mIou mPA values compared previous methods.