EEG signal processing 脑电信号处理方法算法

EEGSIGNALPROCESSINGEEGsignalmodelling1Availablefeatures2Classificationalgorithms3IndependentComponentAnalysis4CONTENTSparseRepresentation51EEGsignalmodellingBioelectricity1Signalgenerationsystem2BIOELECTRICITYSIGNALGENERATIONSYSTEMExcitationmodelSIGNALGENERATIONSYSTEMBIOELECTRICITYLinearModelSIGNALGENERATIONSYSTEMBIOELECTRICITYNonlinearModel2AvailablefeaturesBasicfeatures1Modernmethods2TemporalAnalysisSignalSegmentation:labeltheEEGsignalsbysegmentsofsimilarcharacteristics.BASICFEATURESMODERNMETHODSTemporalCriteriaBASICFEATURESMODERNMETHODSFrequencyAnalysisSuboptimalDFT,DCT,DWT;OptimalKLT(Karhunen-Loève)Demerits:completestatisticalinformation,nofastcalculation.BASICFEATURESMODERNMETHODSSignalParameterEstimationARmodel:Merits:OutperformDFTinfrequencyaccuracy.Demerits:sufferfrompoorestimationofparameters.Improvements:accurateorder&coefficients.MODERNMETHODSBASICFEATURESARcoefficientsestimationmethodsYule-Walkeraryule(x,p)Merits:ToeplitzmatrixLevinson-Durbin,fastest!!!Demerits:withwindowbadresolutionofPSDMODERNMETHODSBASICFEATURESARcoefficientsestimationmethodsCovariancemethodarcov(x,p),armcov(x,p)Merits:withoutwindowgoodresolutionofPSDDemerits:slowBurgarburg(x,p)Merits:accurateapproximationofPSDDemerits:lineskewing&splittingMODERNMETHODSBASICFEATURESMODERNMETHODSBASICFEATURESComparisonPrincipalComponentAnalysisUsesameconceptasSVDDecomposedataintouncorrelatedorthogonalcomponentsAutocorrelationmatrixisdiagonalizedEacheigenvectorrepresentsaprincipalcomponentApplicationdecomposition,classification,filtering,denoising,whitening.MODERNMETHODSBASICFEATURES3SparseRepresentationSparseApproximation1SparseDecomposition2Over-completedictionaryatomsHilbertspace:Signal:Error:“Sparse”:lN,satisfylimitederror.SPARSEAPPROXIMATIONSPARSEDECOMPOSITION,1,2,...kDdkKKNyHllrrrIyd(,)inflllyyDyyNHRMajoralgorithms:BasicPursuit,MatchingPursuits,OMPMatchingPursuits(MP):1st:kth:SPARSEDECOMPOSITIONSPARSEAPPROXIMATION0(1,...),,rikiydsupyd001,rryyddRy1(1,2,...),,kkrikkiRydsupRyd10,nnknrrknyRyddRy与正交nrd1kRyK-SVD:trainingdictionaryPotentialapplicationsforEEG:CoefficientsfeaturesERPdetectionAbnormalEEGdetectionClassificationofdifferentstatusofEEGSPARSEDECOMPOSITIONSPARSEAPPROXIMATION4ClassificationalgorithmsCommonmethods1NaïveBayesLDA:LinearDiscriminantAnalysisHMM:HiddenMarkovModellingSVM:SupportVectorMachineK-meansANNs:ArtificialNeuralNetworksFuzzyLogicCOMMONMETHODS5IndependentComponentAnalysisICAapproaches1Application2IndependentComponentAnalysisICAAPPROACHESAPPLICATIONSICAAPPROACHESAPPLICATIONSICAapproaches:FactorizingthejointPDFintoitsmarginalPDFsDecorrelatingsignalsthroughtimeEliminatingtemporalcross-correlationfunctionBSS:BlindSourceSeparationNormalbrainrhythms,event-relatedsourcesArtefactseyemovement&blinking,swallowAPPLICATIONSICAAPPROACHESTHANKS!