基于机器学习的数字信息处理技术研究与实现

南京航空航天大学硕士学位论文基于机器学习的数字信息处理技术研究与实现姓名：浦海晨申请学位级别：硕士专业：测试计量技术及仪器指导教师：万晓冬20060201I(DSP)DSPDSPDSPBPBPGABPIIABSTRACTThedigitalsignalprocessing(DSP)alreadyobtainedwideapplicationinmanydomainsinthepastseveraldozensyears.ButtheDSPtechnologyalsohasverymanylimitationsalongwiththecomputerandtheinformationtechnologydevelopment.ForinstancetheDSPtechnologylackstheabilitytodowhatonewouldlikewhenfacingtherichkindsofdigitalinformation.ThereforepeopleneedtoextendandexpandDSPtechnologyinordertosatisfythenewrequest.IthastheimportantresearchvaluethatpeopleappliesthemachinelearningasonekindofmatureDSPtechnologyindigitalinformationprocessing.Thepapertakeemailtheonekindofcommondigitalinformationmediumsastheresearchobject.Itrealizestheintelligentprocessingofthemaildigitalinformationthroughtheresearchofmaildigitalinformationprocessingbasedonmachinelearning.Itenhancestheclassificationandrecognitionaccuracyofthemaildigitalinformationinordertosatisfytherequestthatthemailgatewayaccuratelyfiltersthespammail.Thispaperanalyzedthedigitalinformationprocessingtechnologyandtheapplicationofmachinelearningindigitalinformationprocessingfirstly.Anditstudiedhowtoapplythemachinelearningtechnologyinmaildigitalinformationprocessing.Thenitstudiedanddesignedthefeatureselectionalgorithmandtheclassifyalgorithmofmaildigitalinformation.Maildigitalinformationfeaturespacedimensionisveryhuge.Inordertoreducethefeaturedimensionandcausedthemachinelearningalgorithmtobefeasible,ithasutilizedfeatureselectionalgorithmbasedonthefeatureselectionmeasurefunction.Andonthefoundationithasresearchedanddesignedthefeatureselectionalgorithmwhichcombinesthegeneticalgorithmsandthefeatureselectionmeasurefunction.Atthesametime,ithasdesignedtheBPnervenetworkclassifierinordertorealizeintelligentprocessingofmaildigitalinformationandenhancestheclassifiedrecognitioneffect.AnditoptimizedtheBPneuralnetworkbygeneticalgorithmsandhasrealizedtheGABPnetworkclassifier.Finally,thepaperdesignedandrealizedanemailinformationIIIprocessingsystembasedonmachinelearning.Andthesystemwasexperimentedandanalyzedbyemailexamplesandhasgotsomesatisfactionresults.Theexperimentshowedthattheemaildigitalinformationprocessingbasedonmachinelearningtechnologyisfeasibleandeffective.Keywords:digitalinformation,machinelearning,vectorspacemodel,featureselection,geneticalgorithms,artificialneuralnetwork,()VII2.1.........................................................................................113.1VSM..........................................183.3............................283.4....................................................................293.5................333.6............................................................................................353.7............364.1............................................................................................394.2BP.........................................................................404.3BP...........................414.4BP.....................................................434.5BP.................................464.6GABP........................................................................484.7BP.....................................................525.1............................................545.2....................................................565.3............................................575.4........................................................................................585.5.............................................................595.6............................................595.7............................................................605.8....................605.9BP.............615.10BP.............................................................................645.11Linux.................645.12..................................655.13GABP..............................665.14BP.......................................................................675.15BP...................................................................67VIII5.16GABP......................................685.17GABP..................................685.1BP..................................655.2GABP............................69IXDSPDigitalSignalProcessingDIPDigitalInformationProcessingVSMVectorSpaceModelTFTermFrequencyIDFInverseDocumentFrequencyESExhaustiveSearchBIFBestIndividualFeaturesIGInformationGainMIMutualInformationCHIChi-squareStatistic2χ-CECrossEntropyGAGeneticAlgorithmANNArtificialNeuralNetworkBPBack-Propagation11.1(DigitalInformationProcessingDIP)DigitalSignalProcessingDSP1)2)3)21.220505060708020801993(InformationSuper-highway)19951998131998612080[1]1989WaibelLee1989PomerleauCooper199720001.34BPBPGABP51.4BPBPBP62.12.1.160(DigitalSignalProcessingDSP)[2][3]601965FFT60DSP71975A.V.OppenheimR.W.SchaferDigitalSignalProcessingL.R.RabinerB.GoldTheoryandAppeicationofDigitalSignalProcessing[4]80LSIDSPDSP2.1.2(N.Wiener)1948[5](DigitalInformation)(DigitalInformationProcessingDIP)8[6]8020801993(InformationSuper-highway)199819986192.1.32.210[7](Machi