EEG-signal-processing-腦電信號處理方法算法

1、EEG SIGNAL PROCESSING第一頁，共二十七頁。EEG signal modelling1Available features2Classification algorithms3Independent Component Analysis4ContentSparse Representation5第二頁，共二十七頁。1EEG signal modellingBioelectricity 1Signal generation system2第三頁，共二十七頁。 bioelectricitySignal generation systemExcitation model第四頁，共二

2、十七頁。 signal generation systembioelectricityLinear Model第五頁，共二十七頁。 signal generation systembioelectricityNonlinear Model第六頁，共二十七頁。2Available featuresBasic features1Modern methods2第七頁，共二十七頁。Temporal Analysis Signal Segmentation: label the EEG signals by segments of similar characteristics. basic featu

3、resModern methods第八頁，共二十七頁。Temporal Criteria basic featuresModern methods第九頁，共二十七頁。Frequency AnalysisSuboptimal DFT, DCT, DWT;Optimal KLT (Karhunen-Love)Demerits: complete statistical information, no fast calculation. basic featuresModern methods第十頁，共二十七頁。Signal Parameter Estimation AR model: Merits

4、: Outperform DFT in frequency accuracy. Demerits: suffer from poor estimation of parameters. Improvements: accurate order & coefficients. modern methodsBasic features第十一頁，共二十七頁。AR coefficients estimation methodsYule-Walker aryule(x,p) Merits: Toeplitz matrix Levinson-Durbin, fastest! Demerits: with

5、window bad resolution of PSD modern methodsBasic features第十二頁，共二十七頁。AR coefficients estimation methodsCovariance method arcov(x,p), armcov(x,p) Merits: without window good resolution of PSD Demerits: slowBurg arburg(x,p) Merits: accurate approximation of PSD Demerits: line skewing & splitting modern

6、 methodsBasic features第十三頁，共二十七頁。 modern methodsBasic featuresComparison 第十四頁，共二十七頁。Principal Component AnalysisUse same concept as SVDDecompose data into uncorrelated orthogonal componentsAutocorrelation matrix is diagonalizedEach eigenvector represents a principal componentApplication decompositio

7、n, classification, filtering, denoising, whitening. modern methodsBasic features第十五頁，共二十七頁。3Sparse RepresentationSparse Approximation1Sparse Decomposition 2第十六頁，共二十七頁。Over-complete dictionary atomsHilbert space : Signal: Error: “Sparse: lN , satisfy limited error . sparse approximationSparse decompo

8、sition第十七頁，共二十七頁。Major algorithms: Basic Pursuit, Matching Pursuits, OMPMatching Pursuits (MP):1st: kth : sparse decompositionSparse approximation 與 正交第十八頁，共二十七頁。K-SVD: training dictionaryPotential applications for EEG:Coefficients featuresERP detectionAbnormal EEG detectionClassification of differe

9、nt status of EEG sparse decompositionSparse approximation第十九頁，共二十七頁。4Classification algorithmsCommon methods1第二十頁，共二十七頁。Nave BayesLDA: Linear Discriminant AnalysisHMM: Hidden Markov ModellingSVM: Support Vector MachineK-meansANNs: Artificial Neural NetworksFuzzy LogicCommon methods第二十一頁，共二十七頁。5Indep

10、endent Component AnalysisICA approaches1Application 2第二十二頁，共二十七頁。Independent Component Analysis ica approaches applications第二十三頁，共二十七頁。 ica approaches applicationsICA approaches:Factorizing the joint PDF into its marginal PDFsDecorrelating signals through timeEliminating temporal cross-correlation function第二十四頁，共二十七頁。BSS: Blind Source SeparationNormal brain rhythms, event-related sources Artefacts eye movement & blinking, swallow applicationsIca approaches第二十五頁，共二十七頁。THANKS!第二十六頁，共二十七頁。內(nèi)容總結(jié)EEG SIGNAL PROCESSING。Optimal KLT