An Adaptive Feature Extraction and Classification Method of Motion Imagery EEG Based on Virtual Reality

With the aim to solve the problems such as low classification accuracy and weak anti-disturbances in brain–computer interfaces (BCI) of motion imagery, a new method for recognition of electroencephalography (EEG) was proposed in this work, which combined the wavelet packet transform and BP neural network. First, EEG is decomposed by wavelet packet analysis. Then, distance criterion is selected to measure the separable value of the feature frequency bands. Furthermore, the optimal basis of wavelet packet is attained by using a fast search strategy of “from the bottom to the top, from left to right.” The classification feature is extracted by choosing the part wavelet package coefficient, which can attain higher classification evaluation value according to the optimal basis of wavelet packet. And then, the optimal bands are combined with BP neural network. The experimental results show that the proposed method can choose the feature bands of EEGs adaptively, and the highest classification accuracy is 94 %. The correctness and validity of the proposed method is proved. Lastly, establish the virtual robot in MATLAB and use the classification results to control the robot’s arm motion.