Research On The Regularized Extraction Method For Residential Area In Remote Sensing Image

The extraction of feature elements from remote sensing images is one of the key steps of remote sensing mapping,the accuracy of feature extraction directly affects the quality of the mapping results.Nowadays,the remote sensing image semantic segmentation method based on deep learning has become a mainstream method for rapid extraction of terrain features.Residential area is a typical surface feature in remote sensing image,and it is necessary to research on the applicable regularized extraction method,which will provide an effective way to solve the problem of extraction of surface feature in remote sensing mapping.To solve the problem of low accuracy and inaccurate contours of residential area elements in remote sensing images obtained through semantic segmentation network,a multi-feature enhancement Deep Labv3+ is proposed,which enhances the network’s utilization of image feature information by adding a multi-feature enhancement module,and improves the accuracy of network extraction.To solve the problem that the results of semantic segmentation of residential areas cannot be directly applied to remote sensing maps,a post-processing regularization method based on gradual optimization is proposed,which realizes the regularization of residential area extraction results.To solve the problem of cumbersome steps and poor integrity of the post-processing regularization method,a regularized extraction method based on contour regression is proposed,by adding a contour regression module,the end-to-end regularized extraction of residential areas is realized.The experimental results show that compared to the original Deep Labv3+ network,the m Io U of the constructed data set by the multi-feature enhancement Deep Labv3+ is increased by 0.98%,and the extraction of residential area contours is more accurate and complete.Post-processing regularization method based on gradual optimization can process the residential area extraction results into regularized vector data,but the accuracy is limited by the result of semantic segmentation.The regularized extraction method based on the contour regression module has a better effect on the extraction of elements of residential areas with regular shapes,but it is difficult to accurately extract residential areas with large areas and complex outlines.