支持向量机及其在纹理分类中应用

《支持向量机及其在纹理分类中应用》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、II汕头大学硕士学位论文：支持向量机及其在纹理分类中的应用于不同纹理特征的分析，表明当使用支持向量机作为分类器时，在同系列的特征中选择维数更高的特征通常能够得到更好的结果。²将支持向量机与传统的神经网络、LVQ等分类方法在纹理分类上的应用做了比较分析。分类结果表明，支持向量机具有更好更稳定的分类性能，适合于解决高维的纹理分类问题。关键词：纹理分类；支持向量机；统计学习理论；核函数；参数选择ABSTRACTIIIAbstractMachineLearning(ML)basedondataisanimportantresearch

2、contentofpresentintelligenttechnology.Thegoalofmachinelearningisto¯ndtheinternalrelation-shipbetweendatabylearning"thecollecteddata,whichcanbeusedtopredictanddecideonneworunknowndata.TraditionallearningapproachesareusuallybasedontheprincipleofEmpiricalRiskMinimizat

3、ioninsteadofExpectedRiskMinimiza-tion,whichassumethatthetrainingsamplenumbertendsin¯nity.However,theassumptiondoesnotconformtothepracticalapplication,someexcellentapproachintheoryperformnotwellinpractical.StatisticalLearningTheory(SLT)focusesonthemachinelearningofsm

4、allsam-plesizeandcantradeo®betweenthecomplexityofmodelsandgeneralizationperfor-mance,whichincludeVCdimensiontheoryandStructuralRiskMinimizationtheory.SupportVectorMachine(SVM)basedonSLThasexcellentperformanceonsolvingsmallsamplesize,highdimension,andnonlinearpattern

5、classi¯cationproblems,italsosolvee®ectivelytheproblemofover-¯ttingandcurseofdimensionality,andhasgoodgenealizationperformance.Thisthesisaddressonboththetheoryandappli-cationintextureclassi¯cationofSupportVectorMachine.Thecontributionsofthisthesisinclude:²Anoverviewo

6、fboththeoreticalbasisandprincipleofSupportVectorMachineisgiven.Theimplementationalgorithmsareconcerned,andtheirsadvantagesanddisadvantagesandapplicationscopeareshowed.²Summarizetheconstructionapproachesofkernelfunction,includingapproachesbasedonfeaturetransformation

7、,thepropertiesofMercerkernelfunctionandpriorknowledge.Theautocorrelationkernelfunctionhasalsobeenconstructed,andaselectionprocedureofkernelfunctionispresented.IV汕头大学硕士学位论文：支持向量机及其在纹理分类中的应用²AnimprovedgridsearchapproachispresentedforparameterselectionofSup-portVectorM

8、achine.²AnoverallapplicationframeworkforSupportVectorMachineisproposedbasedonthekernelselectionandparametersearchstrategy.²Thispaperalsous