支持向量机若干基础研究及其在图像识别中的应用

《支持向量机若干基础研究及其在图像识别中的应用》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、分类号：至望Q!罾级：垒噩UDC：——学号：!191东J’‘-、博士南大学位掌re．文支持向量机若干基础研究及其在图像识别中的应用研究生姓名：业宁导师姓名：董逸生教授申请学位级别熊±学科专业名称i±篡扭应旦撞苤论文提交曰期垫箜生!星旦论文答辩日期一学位授予单位盔直塞堂学位授予日期答辩委员会主席——评阅人2005年12月ABSTRACTSupporcVectorMachine(SVM)isattractingextensiveaRentionasanovelmethodindatamining·ItisbasedOBstricttheorya

2、ndCansolvetheproblemofovercominglocalsolutionandoverfittingproblemofotheralgorithms．Alsoithasverygoodgeneralizationcapacity．Asaresult，itCanbeappliedinareassuchasfingerprint,facerecognition,DNAdetection,textclassification，OCR，handwriting，diseasedetectionandfailuredetectioni

3、nauto-controlledequipment．Onthebaseofthestatisticleaningtheory,thispaperinvestigatesontheissuesofmodelconstruction，fasttrainingalgorithms，kernelfunctionconstructionandregressionalgorithmsofSVM．Thenasavalidateprocess，weapplytheSVMmethodtothedefectsrecognitioninwoodimagesand

4、theresultispositive·Themainworkofthispaperincludes：1．AMulti—LagrangemultiplierSupporIVectorMachinefasttrainingMethod(MLSVM)basedonthecoordinatedoptimizationofmulti—Lagrangemultipliersisproposedandfourindividualalgorithms，MLSVMl，MLSVM2，MLSVM3andMLSVM4arepresented．Testedwith

5、thestandardtestdatasetsofAdulLWebandMNIST,MLSVM3performsfasterthantheSMOalgorithmwithanimprovementof7．4％to4130％andMLSVM4performsfasterthantheSMOwithanimprovementof300％to4200％．2．Anovelkernelfunctionthatgivesattentiontothesimilarityofbothinputspaceandfeaturespaceisproposed．W

6、ealsopresentanovelorthogonalchebyshevkernelfunction．ThesefunctionsperformwelIintestswithtestdatasets．3．Asupportvectorregressionmethodbasedonclassificationispresentedtosolvethenonlinearregressionproblemwithunknowndatadistributionandmathematicalmodel．4．TheSVMmethodareapplied

7、inidentifyingintemallogdefectsusingCTImageryanditCanautomaticallyfindthedefectsofknotsplitsanddecayNextthe3Dvisualimagesoflogwillbereconstructedincomputersandthe3DdefectswillberecognizedusingSVM．Keywords：SupportVectorMachine,StructuralRiskMinimization，KernelFunction，Classi

8、fication,Regression，PatternRecognition,Log。_。_。-_____-___I_-__·-。______。__-·-。_。-__●-··。_