Camouflage target detection method and system based on pyramid type visual Transform

The invention discloses a camouflage target detection method and system based on pyramid type visual Transform, and belongs to the technical field of computer vision. The method is used for carrying out positioning and pixel-level segmentation on the camouflage object with visual features similar to those of the environment in the image. An input image passes through an encoder based on a pyramid type visual Transform, and image features used for a camouflage target detection task are obtained. The decoder is composed of a segmentation reasoning module based on two-dimensional multi-head self-attention and an edge reasoning module based on one-dimensional multi-head self-attention, and the image features are further analyzed into segmentation masks to serve as an output result of the model. The method can accurately segment the camouflage target with texture deception and edge fuzziness, and effectively solves the problem that a conventional target segmentation method is inaccurate in segmentation of the camo