Abstract: Aimed at the problem that the traditional image segmentation model based on deep learning has low segmentation accuracy for foot ulcer wound area, a segmentation method based on improved UNet is proposed. This method integrated the improved DenseNet network and ASPP into the UNet network, reduced the network parameters and suppressed the interference of irrelevant features on the network model, and extracted multi-scale features of foot ulcer wounds. At the same time, the edge loss function was introduced to solve the problem of poor segmentation ability of model edge details. The experimental results on the DFUC2022 dataset show that the four evaluation indicators of the algorithm, Precision, Recall, MIoU and F2-score, have reached 0.904, 0.915, 0.858 and 0.913 respectively, which are better than the other four segmentation methods.