Remote sensing image change detection method under low registration precision based on graph matching model

Abstract

The invention discloses a remote sensing image change detection method under low registration precision based on a graph matching model. The method comprises the following steps: firstly, carrying outbuilding segmentation by using a multi-scale adaptive convolution remote sensing image semantic segmentation model to obtain segmentation results of different time phases in the same region, namely binary images; in order to solve the problem of region connection after segmentation of a plurality of different buildings, adopting plaque extraction based on an instance segmentation method, so thatdifferent plaques are distinguished, and regions corresponding to the plurality of buildings in different time phases are obtained; carrying out patch matching based on a graph matching model according to the position information corresponding to the patch and other characteristics; and finally, extracting corresponding change information according to the matched information. According to the invention, high-precision change detection of remote sensing images with different time phases under low registration precision is realized. The change detection scheme provided by the invention does notdepend on accurate image preprocessing, especially image correction and registration, and the requirements for image correction and registration are reduced.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a remote sensing image change detection method. Background technique [0002] For remote sensing image change detection, the patent "a method for detecting changes in remote sensing image buildings based on convolutional neural network, CN110136170A" discloses a method for detecting changes in remote sensing images based on deep learning. This patent uses a deep learning network model to learn various collected samples, train and build a semantic segmentation model, obtain a segmented binary image, and then obtain the result of transformation detection. The method described in the patent only designs a new model in terms of building segmentation, but it does not solve the dependence of transformation detection on image preprocessing, especially image correction and registration. The designed segmentation model is only some deformation of the Mask RCNN versio...

AI Technical Summary

Problems solved by technology

In this method, the multi-scale self-adaptive convolution remote sensing image semantic segmentation model is used to segment buildings, and the segmentation results of different time phases in the same area are obtained, that is, binary images; in order to solve the problem of connected areas after segmentation of multiple different buildings, The patch extraction based on the instance segmentation method is used to distinguish different patches and obtain the corresponding areas of multiple buildings in different phases; then, according to the location information and other features corresponding to the patches, the patch based on the graph matching model is carried out. Block matching; finally, according to the matching information, extract the corresponding change information

Images