Classification of Left and Right Hand Motor Imagery Tasks Based on EEG Frequency Component Selection

In this paper, a method based on the time-frequency analysis of EEG frequency spectral Fisher-ratio is proposed to pre-select the most relevant movement-related EEG features. Within this method, combining EEG spectral time-frequency distribution with Fisher criterion, the detailed separability information of the frequency components between two classes of EEG patterns in time-frequency plane over C3, C4, Cz are well characterized, which provides a good guide for selecting the most relevant EEG frequency components. According to Fisher-ratio distribution of EEG spectrum by Matching Pursuit (MP) with high frequency resolution, the matched method Morlet wavelet filter is applied to extract the most relevant EEG frequency components. With the optimized EEG features, two classes of EEG patterns during left and right hand motor imagery are discriminated. Here, BCI competition data are analyzed offline and the satisfactory classification results are obtained, which verify the effectiveness of the proposed method in selecting the most relevant EEG spectral components.