A Semi-supervised Road Segmentation Method for Remote Sensing Image Based on SegFormer

Aiming at the problem that that pixel-level annotations of remote sensing images are difficult to obtain, a semi-supervised road segmentation method for remote sensing images is proposed. Firstly, an unsupervised network is designed to generate pseudo-labels of road images. In this module, a super-pixel segmentation method is used to pre-segment roads in remote sensing images, and then a lightweight convolutional neural network is used to extract road feature information, and to optimize the super-pixel segmentation result to generate the pseudo-label images. Secondly, the loss function of SegFomer is improved to solve the problem that, the difference between the number of front and rear pixels in the remote sensing road image is difficult to accurately segment. Finally, the pseudo-label image and the original image are combined and input to the improved SegFormer network for training. The experiment results show that, the segmentation effect of the proposed method is better than PSPNet, HRNet and other methods.