Remotesensingimagesshowaverypromisingperspectivefordistinguishingtreespecies,especiallythosewiththeveryhighresolutionrangingfrom1to4m.However,thetraditionalmethodologyforclassifyinglandcovertypes,solelydependingonspectralfeatures,whiletextureandotherspatialinformationareneglected,hastheweaknesssuchasinadequatelyutilizationofinformation,lowaccuraciesofclassification,etc.Consideringtothetexturedifferencesamongforestspecies,itismoreimportantforspatialinformationdescriptionofhigh-resolutionremotesensingimagetoimprovetheprecisionoftexturalfeatureschoosing.Inthisstudy,thefactorstoinfluencetheninetexturalfeatureschoosingwereanalyzedandtheresultsshowedthatthemovingwindowsizewasthemainfactortoaffecttheobtainingprocessesoftexturalfeaturesbasedonthegraylevelco-occurrencematrix(GLCM)method,andtheimagerywasthenclassifiedcombiningthemaximumlikelihoodclassification(MLC)methodwiththeoriginalspectralvaluesandtexturefeatures.First,thisstudyutilizedacorrelationanalysisoftheimagesfromaprincipalcomponentanalysis.Second,throughmultipleinformationsources,includingtextualfeaturesderivedfromthedata.Forthehigh-resolutionremotesensingimage,themostpropermovingwindowsizewasdeterminedfrom3×3to31×31.Classificationofthemajortreespeciesthroughoutthestudyarea(theSunYat-SenMausoleuminNanjing)wasundertakenusingtheMLC.Third,toaidforestresearch,classificationaccuracywasimprovedusingtheGLCM.Accordingtocorrelationsamongtexturesandrichnessofthedata,GLCMprovidedthebestwindowsizeandtexturalparameters.Resultsindicatedthatthetexturecharacteristicswereaddinthespectralcharacteristicstoimprovetheprecisionoftheresultsoftheclassification,19×19windowforbestwindow.Thetotalprecisioncanreach66.3226%,Kappacoefficientis0.5840.Eachtreespecieshasgreatlyimprovedaccuraciesoftheclassification.Bythecal
