Feature Extraction for Visual Evoked EEG Based on Wavelet and Bispectrum Analysis
Peng Li · Ceshi jishu xuebao · 2012
Considering that the electroencephalogram(EEG) signal is nonlinear,nonstationary and non-Gaussian,a wavelet-bispectrum analysis method was proposed to extract the feature for visual evoked EEG.Visual evoked EEG data were acquired by Oddball experimental paradigm.First,a coherence average was used to eliminate spontaneous EEG.Then,wavelet decomposition and reconstruction were performed by selecting the appropriate wavelet function and decomposing levels,and the coefficients were whitened for the reconstructed single.Finally,the feature for visual evoked EEG was extracted by bispectrum analysis.The results have shown that this method can obtain the plentiful high-order time-frequency information in the EEG,and has strong advantage for the non-linear processing and Gaussian noise suppression of EEG.