Earth observation image semantic segmentation method based on self-supervised learning

Abstract

The invention provides an earth observation image semantic segmentation method based on self-supervised learning. A semantic image restoration task is used as an auxiliary task of self-supervised learning to pre-train a coding and decoding image restoration network, and saliency detection and an attention mechanism are introduced to improve the regional feature extraction capability of the network, then, the pre-trained network is used for a semantic segmentation task through fine adjustment, and semantic segmentation of the earth observation image data set with only a small number of labels is achieved. According to the method, the problems that an existing semantic segmentation method is difficult to extract features from the earth observation image and the segmentation precision is insufficient under the condition of insufficient annotation data can be solved, and higher semantic segmentation precision is achieved.

Description

technical field [0001] The invention belongs to the technical field of computer vision and semantic segmentation, in particular to a method for semantic segmentation of earth observation images based on self-supervised learning. Background technique [0002] Semantic segmentation is an important and challenging task in the field of computer vision. Its goal is to segment an image into regional blocks of different semantic categories at the pixel level, and give each pixel a category label. Earth observation images are high-altitude images captured by imaging satellites, UAVs, etc. In recent years, the semantic segmentation of earth observation images has attracted widespread attention from scholars, and has been used in crop yield prediction, road network extraction, scene analysis and vegetation coverage. It has important application prospects in many applications. At present, the semantic segmentation of earth observation images has the following problems: (1) the similar...

AI Technical Summary

Problems solved by technology

The present invention can solve the problem that the existing semantic segmentation method is difficult to extract features from the earth observation image and the segmentation accuracy is insufficient in the case of insufficient annotation data, and has higher semantic segmentation accuracy

Images