High-resolution remote sensing image weak supervision building extraction method

Abstract

The invention provides a high-resolution remote sensing image weak supervision building extraction method. The method comprises the steps of class activation image iterative optimization based on an iterative adversarial rising strategy, class activation image inter-pixel relation mining, class boundary detection, building pseudo label generation, semantic segmentation network training and building area extraction. Compared with an existing image-level weak supervision semantic segmentation method, the method has the advantages that an iterative adversarial rising strategy is used, a better building class activation graph can be generated, full mining of class activation graph information is achieved under the condition that extra supervision information is not introduced, and the class equivalence relation between pixels is obtained. Besides, the segmentation performance of the network on boundary processing is further improved by using a gated convolutional layer, so that an object region can be effectively expanded and covered in the boundary, and a high-quality building pseudo tag is generated, thereby enabling the semantic segmentation model to generate a building region with high accuracy and complete segmentation boundary.

Description

technical field [0001] The invention relates to the field of remote sensing, in particular to a method for extracting buildings with weak supervision from high-resolution remote sensing images. Background technique [0002] With the development of satellite remote sensing and aerial photography technology, people can obtain various high-resolution images faster and cheaper. Remote sensing imagery collected from aerial and satellite platforms can be used in a wide variety of applications, such as land-use mapping, urban resource management, and disaster monitoring. Geographic object-based image analysis (GEOBIA) is the main method to extract buildings from high-resolution remote sensing images, but it is difficult to determine the optimal image segmentation scale, and it often requires strong domain expert knowledge when extracting features. Semantic segmentation of high-resolution remote sensing images aims to assign a geographic label to each pixel through an end-to-end me...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the technical problems of difficult label acquisition, incomplete class activation map coverage, and incomplete segmentation boundaries in traditional building extraction methods, the present invention proposes a weakly supervised building extraction method for high-resolution remote sensing images, which focuses on mining between pixels The class equivalence relation and boundary optimization of

Images