摘要
IthaslongbeenacknowledgedthatGISdatacanbeusedasauxiliaryinformationtoimproveremotesensingimageclassification.Inpreviousstudies,GISdatawereoftenusedintrainingareaselectionandpostprocessingofclassificationresultoractedasadditionalbands.Generally,itisfulfilledinastatisticalorinteractivemanner,soitisdifficulttousetheauxiliarydataautomaticallyandintelligently. Furthermore,iftheclassifierrequestscertainstatisticalcharacteristics,theadditionalbandmethodcannotbeusedbecausemostauxiliarydatadonotmeettherequirementsofstatisticalcharacteristics.Ontheotherhand,expertsystemtechniqueswereincorporatedinremotesensingimageclassificationtomakeuseofdomainknowledgeandlogicalreasoning.Butbuildinganimageclassificationexpertsystemwasverydifficultbecauseofthe“knowledgeacquisitionbottleneck”. Spatialdataminingandknowledgediscovery(SDMKD),istheextractionofimplicit,interestingspatialornon_spatialpatternsandgeneralcharacteristics.Weproposedatheoreticalandtechnicalframeworkofspatialdataminingandknowledgediscovery(Lietal.,1997).Andspatialdataminingissupposedtobeusedintwoaspects,oneisintelligentanalysisofGISdata,theotheristosupportknowledgedriveninterpretationandanalysisofremotesensingimages.SDMKDprovidesanewwayofknowledgeacquisitionforremotesensingimageclassification.Severalresearchershavedonesomeworkinthisfield.Eklundetal.(1998)extractedknowledgefromTMimagesandgeographicdatainsoilsalinityanalysisusinginductivelearningalgorithmC4.5.Huangetal.(1997)extractedknowledgefromGISdataandSPOTmultispectralimageinwetlandclassificationusingC4.5too.Inthesetwostudies,geographicdatawereconvertedfromvectortorasterformatinwhichthesamplingsizeisequaltoimagepixelsize.Theimplementationofdataminingtechniquesinspatialdatabase,especiallyinductivelearningmethod,andthecombinationo
出版日期
2000年04月14日（中国Betway体育网页登陆平台首次上网日期，不代表论文的发表时间）
