Remote sensing image target extraction method fusing self-learning semantic features and design features

Abstract

The invention discloses a remote sensing image target extraction method fusing self-learning semantic features and design features, and the method comprises the following steps: extracting internal edge points of a remote sensing image through employing an artificially designed edge operator, completing the initial segmentation of the image according to the edge points, and marking a segmentationobject; learning and extracting building semantic features through an improved Mask R-CNN model, and extracting a building mask image according to the self-learning semantic features; and fusing the remote sensing image segmentation image based on the edge operator and the mask image to obtain a final building extraction image. According to the method, building extraction is completed from two perspectives of self-learning semantic features and artificial design features of the building. According to the model, the problems of wrong target extraction and missing extraction caused by difficultyin traditional artificial feature design can be solved through self-learning semantic features, and the problems of poor edge fitting and local missing of a building extraction result caused by the self-learning semantic features can be perfected through the design of the artificial features.

Description

technical field [0001] The invention relates to a high-resolution remote sensing image target extraction technology, in particular to a remote sensing image target extraction method integrating self-learning semantic features and artificial design features. Background technique [0002] With the advancement of remote sensing satellite technology and the needs of urbanization development, automatic and accurate extraction of building targets in remote sensing images has become an important research direction in the field of digital mapping. At present, the method of extracting target buildings from remote sensing images can be attributed to the extraction method based on artificial design features and emerging deep learning technology. [0003] At present, the mainstream method for extracting target buildings from remote sensing images can be attributed to the extraction method based on artificially designed features. On this basis, according to the principle of building ext...

AI Technical Summary

Problems solved by technology

[0004] Purpose of the invention: In view of the advantages and disadvantages of deep learning and feature extraction, in order to accurately and effectively extract buildings, the present invention provides a remote sensing image target extraction method that combines self-learning semantic features and artificial design features, which overcomes

Images