Unsupervised semantic segmentation method for cross-domain remote sensing image

Abstract

The invention discloses an unsupervised semantic segmentation method for a cross-domain remote sensing image. The method comprises the following steps: acquiring an unlabeled target domain remote sensing image to be segmented; inputting the unlabeled target domain remote sensing image to be segmented into an unsupervised semantic segmentation model which is trained in advance, wherein the unsupervised semantic segmentation model comprises a geometric consistency constraint module, a domain adaptation network module and a semantic segmentation network module; and outputting a segmentation result image consistent with the unlabeled target domain remote sensing image to be segmented in size. According to the method, the segmentation performance of a segmentation model trained on labeled source domain data on a target domain can be improved, so that the dependence of a semantic segmentation task on large-scale labeled data is reduced, and meanwhile, the generalization performance of the semantic segmentation model on different image domains is improved; the method can achieve the accurate segmentation of the to-be-segmented unlabeled target domain remote sensing image, and is higher in segmentation precision.

Description

technical field [0001] The invention belongs to the field of digital image processing, relates to remote sensing image interpretation technology, in particular to an unsupervised semantic segmentation method for cross-domain remote sensing images. Background technique [0002] The semantic segmentation task is to assign a label to each pixel in the image to achieve pixel-level classification of the image content. However, collecting expert-labeled datasets, especially pixel-level annotations, is a labor-intensive process. At present, the common solution in academia is to adapt the source domain and the target domain, so that the model trained on the labeled source domain can be migrated to the unlabeled target domain and achieve acceptable segmentation performance. [0003] In the prior art, domain adaptation methods are usually constructed for general datasets, such as Cityscapes, a real street view dataset for autonomous driving, as the target domain, and street view data...

AI Technical Summary

Problems solved by technology

Based on this property, the present invention designs a geometric consistency constraint to improve the adaptation effect of the semantic segmentation model in remote sensing images from the source domain to the target domain, and proposes an unsupervised semantic segmentation method for cross-domain remote sensing images, which can Solve the problem of low segmentation accuracy of the traditional model trained on the source domain on the target domain

Images