多尺度分割原理与应用.ppt

《多尺度分割原理与应用.ppt》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、HierarchicalSegmentationReadingReportHSegCONTENTSPARTONEAbouttheIntroduction.PARTTWOTheHsegSegmentation.PARTTHREEApplicationinENVI.1PARTONEAbouttheIntroduction.AbstractRecentadvancesinspectral–spatialclassificationofhyperspectralimagesarepresentedinthispaper.Severaltechniquesareinvestigatedforcom

2、biningbothspatialandspectralinformation.Highlighttheimportanceofspectral–spatialstrategiesfortheaccurateclassificationofhyperspectralimagesandvalidatetheproposedmethods.IntroductionAdvantagesofhyperspectralremotesensortechnology:Thedetailedspectralinformationincreasesthepossibilityofmoreaccuratel

3、ydiscriminatingmaterialsofinterest.Thefinespatialresolutionofthesensorsenablestheanalysisofsmallspatialstructuresintheimage.Manyoperationalimagingsystemsarecurrentlyavailableprovidingalargeamountofimagesforvariousthematicapplications.But,italsobringssomeproblem:theCurseofDimensionalityandtheneedf

4、orspecificspectral–spatialclassifiers.PS:维数灾难(CurseofDimensionality):通常是指在涉及到向量的计算的问题中，随着维数的增加，计算量呈指数倍增长的一种现象。维数灾难在很多学科中都可以碰到，比如动态规划，模式识别和影像识别等。IntroductionTheCurseofDimensionalityofhyperspectralremotesensortechnology:Inhigh-dimensionalspaces,normallydistributeddatahaveatendencytoconcentrateinthe

5、tails,whichseemstobecontradictorywithitsbell-shapeddensityfunction.therateofconvergenceofthestatisticalestimationdecreaseswhenthedimensiongrowswhileconjointlythenumberofparameterstoestimateincreases,makingtheestimationofthemodelparametersverydifficult.withalimitedtrainingset,beyondacertainlimit,

6、theclassificationaccuracyactuallydecreasesasthenumberoffeaturesincreases.IntroductionHowtobuildaccurateclassifiersforhyperspectralimages?SVMsperformanonlinearpixel-wiseclassificationbasedonthefullspectralinformationwhichisrobusttothespectraldimensionofhyperspectralimages.Iterativestatisticalclass

7、ifierbasedonMarkovrandomfield(MRF)modeling.NotethatrecentlyadaptiveMRFhavebeenintroducedinremotesensing.Useadvancedmorphologicalfiltersasanalternativewayofperformingjointclassification.PS:鲁棒性(Robust):即系统的健壮性，是在异常和危险情况下