基于环形卷积的遥感影像语义分割方法

打开文本图片集

关键词：环形卷积；语义分割；遥感影像；离散傅里叶变换

中图分类号：TP391.41 文献标志码：A 文章编号：1001-3695（2025）12-034-3793-06

doi： 10.19734/j.issn. 1001-3695.2025.03.0099

RingNet： semantic segmentation of remote sensing images based on ring convolution

XuRui¹，HanBin¹，ChenFei¹，ZhangZihan² （1.ScolfompueguUesityfece&ZgJua；hlftfall Nanjing Agriculture University，Nanjing210o95，China）

Abstract：Toaddress the limitationsof existing encoder-decoder-based semanticsegmentationalgorithms forremotesensing images，particulalyintersofinsuffcientconextualunderstandingandporsmallojectecognitin，thispaperproposeda novelsemanticsegmentationnetwork caledRingNet，whichwasbuiltuponring-shapedconvolution.Byintroducingringconvolutionlayers toencode semantic features withradial orcircular distributions，RingNet incorporatedtwo key modules：aing residual module（RingRes）andaring-shaped pyramidconvolutionmodule（RingSPP）tocapture multi-scalecontextual information.The networkadoptedResNet18asitsbackboneand integratedtheRingRes moduleintheshalowlayers toexpand the receptive field whilepreservingoriginaltextureinformation.Inthedeeper layers，RingSPPleveragedringconvolutionsof variousradiicombinedwithchannelatentionmechanismstoextractrichsemanticandspatialfeatures.Experimentsconducted on the Potsdam and Vaihingen high-resolution remote sensing datasets demonstrate that RingNet outperforms mainstreamsegmentation models in terms of mean overall accuracy， F1 -score，and mIoU，and achieves superior performance in preserving semanticdetailsandobjectboundaries.Theringconvolution method hasbeen proven to haveanimproved efect inthe semantic segmentation task of remote sensing images.

Key words：ring convolution；semantic segmentation；remote sensing images；discrete Fourier transform

0 引言

高分辨率的遥感图像[1]处理技术，在城市规划[2]、地表分类、环境监测3灾害预警等各个领域的价值日益凸显。（剩余14169字）