Road crack segmentation method based on improved U-Net network

The invention discloses a road crack segmentation method based on an improved U-Net network. The method comprises the following steps: firstly, preprocessing an original road crack data set to obtain a preprocessed road crack data set; then, an improved U-Net road crack segmentation network is constructed; inputting the preprocessed road crack data set into an improved U-Net road crack segmentation network for training to obtain a trained network model; and finally, segmenting the preprocessed to-be-predicted road crack image by using the trained network model to obtain a segmentation result graph of the road crack. Compared with traditional U-Net, the network model is excellent in performance in a plurality of evaluation indexes, the segmentation result is close to the effect after data processing, the robustness is high, the problem that the segmentation precision is reduced due to the fact that the background is complex and interference is large can be effectively solved, and the road crack segmentation ac