The research of brain-computer interface based on AAR parameters and neural networks classifier

The brain-computer interface(BCI) based on motor imagery was investigated in this paper. A neural networks classifier was adopted to solve the problem of lower classification accuracy in BCI. Firstly, mu rhythm EEG was obtained with a bandpass filter from the subject´s scalp electroencephalography (EEG). Secondly, the Kalman Filter algorithm was used to build the adaptive autoregressive model from EEG. The model parameters were used as features of EEG. Lastly, the AAR feature parameters were classified by the neural networks classifier. A compare on the performance between the neural networks and linear discriminant analysis(LDA) was conduct in the simulation. The results show the performance of neural networks is higher than linear discriminant analysis.