支持向量機(jī)若干基礎(chǔ)研究及其在圖像識(shí)別中的應(yīng)用.pdf

1、分類號(hào)：至望Q!罾級(jí)：壘噩UDC：——學(xué)號(hào)：!19111東J’‘、博士南大學(xué)位掌re文支持向量機(jī)若干基礎(chǔ)研究及其在圖像識(shí)別中的應(yīng)用研究生姓名：業(yè)寧導(dǎo)師姓名：董逸生教授申請(qǐng)學(xué)位級(jí)別熊學(xué)科專業(yè)名稱i篡扭應(yīng)旦撞苤論文提交曰期墊箜生!星旦論文答辯日期一學(xué)位授予單位盔直塞堂學(xué)位授予日期答辯委員會(huì)主席——評(píng)閱人2005年12月ABSTRACTSupporcVectorMachine(SVM)isattractingextensiveaRention

2、asanovelmethodindataminingItisbasedOBstricttheoryandCansolvetheproblemofovercominglocalsolutionandoverfittingproblemofotheralgorithmsAlsoithasverygoodgeneralizationcapacityAsaresult，itCanbeappliedinareassuchasfingerprint

3、facerecognitionDNAdetectiontextclassification，OCR，handwriting，diseasedetectionandfailuredetectioninautocontrolledequipmentOnthebaseofthestatisticleaningtheorythispaperinvestigatesontheissuesofmodelconstruction，fasttraini

4、ngalgorithms，kernelfunctionconstructionandregressionalgorithmsofSVMThenasavalidateprocess，weapplytheSVMmethodtothedefectsrecognitioninwoodimagesandtheresultispositiveThemainworkofthispaperincludes：1AMulti—Lagrangemultipl

5、ierSupporIVectorMachinefasttrainingMethod(MLSVM)basedonthecoordinatedoptimizationofmulti—Lagrangemultipliersisproposedandfourindividualalgorithms，MLSVMl，MLSVM2，MLSVM3andMLSVM4arepresentedTestedwiththestandardtestdatasets

6、ofAdulLWebandMNISTMLSVM3performsfasterthantheSMOalgorithmwithanimprovementof74％to4130％andMLSVM4performsfasterthantheSMOwithanimprovementof300％to4200％2Anovelkernelfunctionthatgivesattentiontothesimilarityofbothinputspacea

7、ndfeaturespaceisproposedWealsopresentanovelorthogonalchebyshevkernelfunctionThesefunctionsperformwelIintestswithtestdatasets3Asupportvectorregressionmethodbasedonclassificationispresentedtosolvethenonlinearregressionprob

8、lemwithunknowndatadistributionandmathematicalmodel4TheSVMmethodareappliedinidentifyingintemallogdefectsusingCTImageryanditCanautomaticallyfindthedefectsofknotsplitsanddecayNextthe3Dvisualimagesoflogwillbereconstructedinc

9、omputersandthe3DdefectswillberecognizedusingSVMKeywords：SupportVectorMachineStructuralRiskMinimization，KernelFunction，ClassificationRegression，PatternRecognitionLog。_。_。________I___。______。__。_。__●。_。。。_。。。。。。___●。。。’’’。