EEG Signals Recognition of Active Brain-Computer Interface Based on Sequential Encoding

Abstract

To complement the existing experimental paradigm of motor imagery, an active brain-computer interface (BCI) experimental paradigm based on the sequential encoding of speech imagery and motor imagery is proposed. At the same time, a sub-time window filter bank common spatial pattern (STWFBCSP) algorithm is proposed to overcome the deficiency of common spatial pattern (CSP). In this algorithm, the imagery period is divided into five overlapping time windows and each time window is filtered by multi-frequency. Feature extraction and selection are achieved by CSP and mutual information, respectively. The final decision output is determined by the voting results of five support vector machines (SVMs). Using STWFBCSP, the average classification accuracy of electroencephalogram (EEG) signals of 12 subjects is 84.87%. Compared with CSP, filter bank common spatial pattern (FBCSP), and common time-frequency-spatial pattern (CTFSP) algorithm, the result of the proposed algorithm is improved by 14.38%, 9.02%, and 3.97%, respectively. The results show that the EEG signals of this experimental paradigm can be better extracted and classified by STWFBCSP, and the practicability of BCI is also improved.