资源描述：
《Multi-Classifier Systems(MCSs)of Remote Sensing Imagery Classification Based on Texture Analysis.pdf》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、Multi-ClassifierSystems(MCSs)ofRemoteSensingImageryClassificationBasedonTextureAnalysisHongfenLi,GuangdaoHu,andJiang-fengLiInstituteofMathematicalandRemoteSensingGeology,ChinaUniversityofGeosciences,Wuhan430074,Chinalihongfen_dida@163.comAbstract.Thisarticleconcernsmethodsofim
2、provingtheaccuracyoflandcovermapsusingVeryHigh-resolutionSatellites(VHRS).Itdiscussestwomethodsforincreasingtheaccuracyofclassifiersusedinlandcovermapping.OneistextureanalysisusingGLCMmethodandtheotherismultipleclassifiersystem(MCSs)usingvotingrules.AcasestudyofQuickBirdImager
3、yofanareainChenggongCountyofYunnanProvinceisconductedbasedonananaly-sisofQuickBirdimagery.Theexperimentresultsshowthatthesetwomethodscanimprovetheaccuracygreatly.Theapplyingoftexturebandsmakesanincreaseof2.6816%,andtheMCSsmakeanincreaseof3.9512%.Keywords:multipleclassifiersyst
4、em(MCSs),co-occurrenceprobability,textureanalysis,land-coverclassification.1IntroductionCurrently,Veryhigh-resolutionsatellites(VHRS)includingQuickBirdprovideausefulwayforustoperiodicallymonitordetailedlanduseinbroadareas.PixelsizesonthegroundofVHRSsensorsaresmallenoughtocaptu
5、regeometricaldetailsofcommonlandusepatches,andmostgeographictexturesonVHRSimages.Comparedwithmedium/lowresolutionimage,VHRSimagerygetsmoretexturecharacteristicsandgeographicstructureinformation.Buttheconventionalclassifica-tionalgorithmsbasedonpixelspectrumoftengetlowaccuracy.
6、Theconventionalcomputerclassificationmethodsfocusonthefollowing6as-pects:1)theclassificationbasedonmathematicstatistics;2)theclassificationbasedonimageryfeatures;3)theclassificationofRSimageryofsinglesourceofsensor;4)theclassificationbasedonsinglepixel;5)hardclassificationthat
7、everypixelmustbelongstoaparticularclass;6)oneclassifierthroughouttheclassification;thelimita-tionofthesetechnologiescannotmakeamoreaccurateresult,newmethodmainlydosomeimprovementonthese6aspects.Inthispaper,twokindsofsolutionsareproposedtoimprovetheclassificationac-curacyofVHRS
8、,1)textureanalysisusingGLCMmethod;2)Multipleclassifiersys-tem
