Research Of Brain-Computer Interface Based On Motor Imagery EEG

Electroencephalogram (EEG) is the overall reflection of brain nerve cells electrophysiological activity on cerebral cortex. Because there are correlation between EEG and conscious state of brain, it is possible to identify different conscious states by classification of different EEG models and then forms a communication control system during the peripheral devices and brain, which is called as brain-computer interface (BCI). The BCI’s essence is in order to achieve the man-computer communication by EEG to infer people’s idea or purpose. Motor EEG is generated by the imagination of limb movements accompany with actual physical action. EEG evoked during motor imagery has the characteristics of event-related desynchronization (ERD) and event-related synchronization (ERS), by analyzing this EEG we can judge the movement intention and control peripheral devices. Therefore, motor EEG is a widely used characteristic signal of BCI system.This thesis discusses the feature extraction and classification algorithm of motor imagery EEG, and understands the current research status of it, the focused on study an algorithm based on wavelet analysis. The following two feature extraction methods were proposed, the feature extraction based on wavelet transform coefficients and coefficient averages and the feature extraction based on wavelet packet transform sub-band coefficient combining with sub-band energies. The selection of classifier is much vital, so we choose the probabilistic neural network (PNN) to achieve signal classification after compared the advantages and disadvantages of current classifiers. Then we will send the features which meet the time-frequency characteristics of motor EEG to PNN classifier. All methods mentioned above are used to process the EEG data offered by University of Technology Graz on2003BCI competition, and preliminary validate the effectiveness of our algorithms. Then we design an experiment which includes two motor EEG missions (imagine and lift the left arm, imagine and lift the right arm) using the Emotiv EPOC system, and analyzing the experiment data by these methods mentioned above. The recognition accuracy is an indictor; we will compare the classification results of the researched pattern recognition method, and pick out the better algorithm. It proved that the feature extraction based on wavelet packet transform sub-band coefficient combining with sub-band energies has good time-frequency resolution and accuracy classification results. Finally, we explored the research of the online and real-time BCI system, and made some progress on the subsequent research.