资源描述：
《面向对象的高分辨率遥感影像城区建筑物分级提取方法》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。
1、第39卷第1期测绘学报Vol.39,No.12010年2月ActaGeodaeticaetCartographicaSinicaFeb.,2010文章编号:1001-1595(2010)01-0039-07面向对象的高分辨率遥感影像城区建筑物分级提取方法1,21,21,21,21,2陶超,谭毅华,蔡华杰,杜博,田金文1.华中科技大学图像识别与人工智能研究所,湖北武汉430074;2.华中科技大学多谱信息处理技术国家重点实验室,湖北武汉430074Object-orientedMethodofHierarchicalUrbanBuildingExtractionfromHigh-
2、resolu-tionRemote-SensingImagery1,21,21,21,21,2TAOChao,TANYihua,CAIHuajie,DUBo,TIANJinwen1.InstituteofPatternRecognitionandArtificialIntelligence,HuazhongUniversityofScienceandTechnology,Wuhan430074,Ch-ina;2.TheStateKeyLaboratoryforMult-ispectraInformationProcessingTechnology,HuazhongUnivers
3、ityofScienceandTech-nology,Wuhan430074,ChinaAbstract:Anautomaticurbanbuildingextractionmethodforhigh-resolutionremote-sensingimagery,whichcom-binesbuildingsegmentationbasedonneighbortotalvariationswithobject-orientedanalysis,ispresentedinthispaper.Aimedatdifferentextractioncomplexityfromvari
4、ousbuildingsinthesegmentedimage,ahierarchicalbuildingextractionstrategywithmult-ifeaturefusionisadopted.Firstly,weextractsomerectanglebuildingswhichremainintactaftersegmentationthroughshapeanalysis.Secondly,inordertoensureeachcandidatebuildingtar-gettobeindependent,multidirectionalmorphologi
5、calroad-filteringalgorithmisdesignedwhichcanseparatebuildingsfromtheneighboringroadswithsimilarspectrum.Finally,wetaketheextractedbuildingsandtheexcludednon-buildingsassamplestoestablishprobabilitymodelrespectively,andBayesiandiscriminatingclassifierisusedformakingjudgmentoftheothercandidate
6、buildingobjectstogettheultimateextractionresult.Theexperimentalresultshaveshownthattheapproachisabletodetectbuildingswithdifferentstructureandspectralfeaturesinthesameimage.Theresultsofper-formanceevaluationalsosupporttherobustnessandprecisionoftheapproachdeveloped.Keywords:high-resolutionre
7、mote-sensingimagery;buildingextraction;objec-torientedmethod;morphology;Bayesianrule摘要:提出一种高空间分辨率遥感影像城区建筑物自动提取方法。该方法将面向对象的思想融入到基于邻域总变分的建筑物分割方法中,并通过分析分割后不同类型建筑物提取的难易程度,提出一种多特征融合的建筑物对象分级提取策略:首先通过形状分析检测一部分分割完整的矩形建筑物目标,然后采用新提出的多方向形态学道路滤波算法将建筑物与邻近光谱相似的道
