Motor imagery EEG signals analysis based on Bayesian network with Gaussian distribution

As a novel communication channel from brain to machine, the research of Brain–computer interfacing has attracted more and more attention recently. In this paper, a novel method based on Bayesian Network is proposed to analyze multi-motor imagery tasks. On the one hand, the information of channels physical positions and motor imagery class information mean value are adopted as constrains in BN structure construction. On the other hand, continuous Gaussian distribution model is used to model the Bayesian network nodes other than discretizing variable in traditional methods, which would reflect the real character of EEG signals. Finally, the network structure and edge inference score are used to construct SVM classifier. Experimental results on the BCI competition dataset BCI IIIa and our own lab collected dataset show that the average accuracy of the two experiments are 93% and 88% based on edge selection, which are better comparing to current methods.