Remote sensing image semantic segmentation method based on class center prototype

The invention discloses a remote sensing image semantic segmentation method based on a class center prototype, and the method comprises the steps: inputting a remote sensing image to be subjected to semantic segmentation into a trained semantic segmentation model, obtaining pixel features through a feature extractor composed of an encoder and a decoder, generating a plurality of final local class prototypes during the training of a classifier module, and carrying out the semantic segmentation of the remote sensing image. And calculating the similarity between the pixel features and the final local class prototypes by using a similarity conversion function, and taking the semantic class to which the final local class prototype with the maximum similarity belongs as a semantic segmentation result of the remote sensing image. According to the method, the semantic segmentation model is constrained by introducing the orthogonal constraint, the first constraint and the second constraint, so that the generation dire