PP-UNet: Improved Road Crack Segmentation Method Based on U-Net

Addressing the issues of the basic U-Net model's insufficient segmentation effect on road cracks, inadequate fine segmentation of crack contours, difficulty in recognizing narrow cracks, and low segmentation accuracy, this paper proposes an improved road crack segmentation method based on U-Net. Firstly, an enhanced encoder is designed to improve feature extraction capabilities. Secondly, Polarized Self-Attention mechanism is introduced to refine the skip connections of U-Net. Finally, Atrous Spatial Pyramid Pooling module is added before feature fusion to obtain the improved model. Experiments conducted on the public dataset Crack500 show that compared to the original network, the improved model's Kappa coefficient, MIoU, and Dice coefficient have increased by 4.79%, 3.26%, and 2.42% respectively. The method proposed in this paper better adapts to the task of road crack segmentation.