EEG Recognition Through Time-varying Vector AR Model

EEG contains a lot of physiological and disease information and can be used for clinical diagnosis by some appropriate treatment. And it is possible to identify the different consciousness through the recognition of EEG signal. The characteristic parameters of the EEG signals can be extracted and the EEG signals can be classified by the computer technology, the non-stationary time series analysis and signal processing methods.In this thesis, the pattern recognition of EEG signals of different movement imagination is studied. After the pretreatment of EEG signals, the most different time series segment between two different patterns is selected. Meanwhile, t he linear regression equation is established for extracting deterministic component and stochastic component.Then, the stochastic component is analyzed by use of TVVAR(Time Var ying Vector Auto-regressive) model to obtain the residuals. For recognizing, the norm and the minimum singular value of the error matrix, the Mahalanobis distance and the of the Mahalanobis distance are used to identify the EEG signals. Through research and analysis, the Mahalanobis distance and the of the Mahalanobis distance are chosen. In addition, the series segments with the length of twelve windows is chosen to classified the EEG signal by using the TVVAR model and the average recognition rates of the Mahalanobis distance and the of the Mahalanobis distance are 96.11%, 95.00% respectively. The simulation result shows that the TVVAR model is reliable and effective to analyze the EEG signals of different movement imagination. The study provides a new way to recognize the EEG signals, and we hope the result is useful in the prosthetic technology to help the disabled individuals to manipulate prostheses effectively by EEG.