Intestinal polyp boundary segmentation method based on sparse connection and global feature enhancement

The invention relates to an intestinal polyp boundary segmentation method based on sparse connection and global feature enhancement, and the method comprises the steps: taking a sparse harmonic connection network as an encoder trunk, extracting features of an input intestinal polyp image, carrying out the parallel sampling of the features of the deepest layer at different sampling rates through a cavity space convolution pyramid pooling module, and carrying out the segmentation of the boundary of the intestinal polyp. A parallel sampling result and the rear two-layer features of the encoder are input into a partial feature decoder together for decoding, and a rough initial prediction map is generated; a shallow layer attention module is used for extracting the rough initial prediction map and shallow layer boundary information in shallow layer features, and the shallow layer boundary information is used for preliminarily refining the boundary of the initial prediction map; deep features are processed by a glo