Analysis and classification of hybrid BCI based on motor imagery and speech imagery

Abstract

As a new communication mode between human and computers, brain-computer interfaces (BCIs) based on electroencephalography (EEG) have been widely studied. In order to increase the operational dimension of BCIs, this paper proposes a hybrid BCI based on motor imagery and speech imagery. According to one versus one calculation model, common spatial pattern (CSP) is extended to a three-category algorithm. The energy eigenvalues extracted by CSP are combined with synchronization eigenvalues, which are respectively calculated by cross-correlation function and phase locking value (PLV). The maximum difference of synchronization between two mental tasks is considered as a method to confirm the channel pairs. The experimental results of ten subjects show that: the highest average classification accuracy in three mental tasks is speech imagery (74.3%); imagining left hand movement is followed (71.4%); imagining right hand movement is the last one (69.8%). The classification results can be improved by combining synchronization and CSP, and the synchronization from cross-correlation function is better than PLV. The frequency range and channel pairs both have certain individual differences. By individually adjusting the appropriate settings for each subject, the usage efficiency of BCIs can be improved.