1. The context information capture module is adopted in the network. By expanding the receptive field, while retaining the detailed information, the multi-scale feature fusion is carried out in the central part, which alleviates the problem of the loss of detailed information in the traditional algorithm; Establish long-distance dependencies between them;
2. Segmentation network: edge extraction network + detail optimization network
E-net
D-not
Concatenate the original image with the 3-channel segmentation result of E-Net as the 6-channel input of D-Net