Study On Preprocessing And Classification Algorithms In Brain Computer Interface

The key technology of brain computer interface (BCI) is to translate the input EEG signal from the user to the control command output. The signal processing of BCI includes three parts:preprocessing, feature extraction and classification. EEG signal collected from the scalp is often contaminated by different artifacts, among which ocular artifacts (OA) are most serious. Specific to this problem, this paper studies the ocular artifacts removal methods based on blind source separation (BSS).First, after proving the validity of independent component analysis (ICA) to remove OA, specific to the two problems of ICA methods, which are the manual identification of artifact component and the spectrum leakage of EEG into artifact component, we combine ICA with wavelet transform (wICA) to remove OA. Experiment results testify that wICA performs better than ICA. Besides, we also turns out that second order statistic identification (SOBI) removes more artifacts and preserves more brain signal than ICA.In order to solve the two problems of OA removal methods based on BSS, making full use of the differences of spatial distribution between EEG and EOG, we applies canonical correlation analysis(CCA) to remove OA in a new way, guaranteeing that the first component computed by CCA is the artifact component. Then wavelet threshold is employed to recover the brain signal leaked in the artifact component. The performance of the proposed method is compared to three popular ocular artifacts removal methods (CCA, SOBI and wavelet-ICA) in terms of correlation coefficient and signal-to-artifact ratio (SAR). It shows that wCCA's performance is better than the other three, removing the most ocular artifacts fromEEG recording automatically without altering the cerebral components.At last, we study the signal processing methods in BCI based on P300 signal. The preprocessing part includes five steps and we extract the time domain waveform of P300 after downsampling as the feature, and then select stepwise linear discriminate analysis (SWDA) as the promising online BCI classifier. Through the standard P300 data of BCI Competition III, we study the influences of the bandpass filtering parameters, electrodes number and averaging number to the result of SWDA classifier.