Optimal EEG feature extraction based on R-square coefficients for motor imagery BCI system

In cue based motor imagery brain computer interface (BCI) paradigm, subject was stimulated by different cue to distinguish different imagination task and proceed by imagination tasks. These evoked the specific related frequency rhythms. But the problem appears as theses rhythms change due subject and imagination task. While some subject is fast induce the electrical signal rhythms, other has long latency in induce them. This related time temporal problem of imagination. To boost the accuracy of EEG signal translation for reliable system, in this paper, we proposed a method for extracting optimal feature. First, EEG signals were extracted and apply to Laplace filter and band pass filter with pass band frequency of 7Hz to 40Hz. Channels C3, Cz, and C4 were applied to Short-Time Fourier Transform with frequency band of 1Hz. R-square correlation coefficient of three channels were found and selected the best frequency and time parameter. Finally, with the selected parameters, optimal STFT feature extracted. We simulated the classification accuracy with linear discriminant analysis with BCI competition IV, III and our laboratory dataset.