资源描述：
《基于非线性光谱混合模型的遥感影像提取城市不透水层_英文_》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40卷第1期光子学报Vol.40No.12011年1月ACTAPHOTONICASINICAJanuary2011ArticleID:10044213(2011)0100136UrbanImperviousSurfaceExtractionfromRemoteSensingImageBasedonNonlinearSpectralMixtureModelXIAJunshi,DUPeijun,CAOWen(KeyLaboratoryforLandEnvironment
2、andDisasterMonitoringStateBureauofSurveyingandMapping,ChinaUniversityofMining&Technology,Xuzhou,Jiangsu221116,China)Abstract:Aimingatovercomingthelimitationsinextractingimpervioussurfacebytraditionalmethods,twononlinearspectralmixturemodels,MixtureTunedMa
3、tchedFiltering(MTMF)andMultiLayerPerceptron(MLP)neuralnetwork,areusedtodecomposeallpixelstothefourfractionimagesrepresentingtheabundanceoffourendmembers.Inthesemodels,MTMFperformsapartialunmixingbyonlyfindingtheabundanceofasingle,userdefinedendmember,b
4、ymaximizingtheresponseoftheendmemberofinterestandminimizingtheresponseofthecompositeunknownbackground.TheMLPisahierarchicalstructureofseveralperceptrons,andcapableoflearningarichvarietyofnonlineardecisionsurfaces.TheMaximumNoiseFraction(MNF)isusedtotransfo
5、rmthesixbandsofTMimageintoanewfeaturespaceandthefirstthreecomponentsaccountingforthemajority(morethan90%)oftotalinformationcontentareusedtoendmemberextraction.Afterthat,thePurePixelIndex(PPI)isusedtoselectpurepixels.TheNdimensionalvisualizerisusedforassis
6、tingselectionoffourendmembers:vegetation,highalbedoobjects,lowalbedoobjectsandsoil.Thefractionimagesarederivedtorepresenttheabundanceoftheabovefourendmember.Impervioussurfaceisestimatedbyanalyzinghighalbedoandlowalbedofractionimages.QuickBirdmultispec
7、tralimageisusedtoevaluatetheaccuracyofimpervioussurfaceextractionbydifferentmethods.Experimentalresultsindicatethattheaccuracyofartificialneuralnetworkishigherthanothers,whichmeansnonlinearspectralmixturemodelsisalsoeffectivetoimperviousareaextraction,eve
8、nbetterthanlinearmodels.Keywords:Spectralmixturemodel;Impervioussurface;Artificialneuralnetwork;MultilayerperceptronCLCN:TP7DocumentCode:Adoi:10.3788/gzxb20114001.00130Introductionlandscape,impervio
