基于超像素的高分辨率影像建筑物提取

《基于超像素的高分辨率影像建筑物提取》由会员上传分享，免费在线阅读，更多相关内容在应用文档-天天文库。

1、基于超像素的高分辨率影像建筑物提取莫中夏张新长张志强中山大学地理科学与规划学院建筑物提取对于基础地理数据获取与更新具有重要意义。针对传统的基于像元方法难以利用影像空间、上下文信息以及面向对象方法最优分割尺度难以确定等问题，该文提出一种基于超像素的高分辨率影像建筑物提取方法。首先利用LiDAR点云数据生成归一化高程模型以获取地物的高度特征，并通过Layerstacking方式与高分辨率影像融合;随后基于SimpleLinearIterativeClustering(SLIC)算法分割高分辨率影像的RGB彩色图像牛.成超像素;继而基于分割所得的超像素中

2、的每个像元对超像素进行特征计算;最后使用基于RBF核的支持向量机方法进行超像素级别分类，得到建筑物提取结果。实验结果表明，该方法能够有效识别出简单场景和复杂场景下的建筑物，显著消除了传统基于像元方法出现的“椒盐效应”，同时避免了面向对象方法中最佳分割尺度选择的难题，取得了较好的建筑物提取效果。关键词：建筑物;超像素;SLIC;LiDAR;基金：国家自然科学基金重点项0(41431178);国家自然科学基金面上项目(41671453)BuildingExtractionBasedonSuperpixelsofHigh-ResolutionImageM

3、OZhong-xiaZHANGXin-changZHANGZhi-qiangSchoolofGeographyandPlanning,SunYat—senUniversity;Abstract：Traditionally,thespaceandcontextinformationaredifficulttobeusedinpixel-basedmethodandtheoptimalsegmentationscaleishardtobedeterminedintheobject-orientedmethod.Tosolvetheproblem,ahi

4、gh-resolutionimagebuildingextractionmethodbasedonsuperpixelwasproposedinthispaper.SimpleLinearIterativeClustering(SLIC)isakindofsuperpixelssegmentationalgorithmwhichhasbeendevelopedinrecentyears.Ithasthefollowingcharacteristics,whichincludessimpleprinciple,fastsegmentation,hig

5、hdegreeofattachmenttotheboundaryofthefeature,andmoreattentiontotheshape.However,therearerelativelyfewstudiesontheuseofSLICinthefieldofremotesensing.Inthisstudy,theRGBcolorimageofthehigh-resolutionimagewassegmentedbasedontheSLICalgorithmtogeneratesuperpixels.Then,thenormalizede

6、levationmodelwasgeneratedbyusingtheLiDARpointclouddatatoobtaintheheightinformationofthefeatures,andthenitwasmergedwiththehigh-resolutionimagebythewayoflayerstacking.Finally,theSupportVectorMachine(SVM)methodbasedonRBF'kernelwasusedtoclassifythesuperpixels,andtheresultofbuildin

7、gextractionwasobtained.Theexperimentalresultsshowedthattheproposedmethodcouldeffectivelyidentifythebuildingsinsimplescenesandcomplexscenes,andeliminatethesaltandpeppereffect"ofthetraditionalpixel-basedmethod.Atthesametime,itavoidedtheproblemoftheoptimalsegmentationscaleselecti

8、onintheobject-orientedmethodandgotabetterbuildingextraction.T