Research On Feature Extraction Algorithms Of Motor Imagery EEG

Study of motor imagery EEG signals is a hot topic in the current, it give a convenient way to communicate with the external environment to people who have severe movement dysfunction. Through the BCI designed, we can successfully “read” the EEG signals. Realizes information exchanged between the brain and the environment, set up a “second way” without nerves and muscles.In order to test the motor imagery EEG signals of right-hand or left-hand better and to research the EEG more convenient, designed a system of EEG collection and processing. Used the VC and Matlab to mix programed, made the EEG signals acquisition and processing integrated. And the judgement of right or left hand come true. The main structure of this EEG system include: hardware and software interface program, user information input and read, electroencephalogram mapping, data save, algorithm selection, use Matlab processing, feedback result, etc.In the Matlab environment, realized two functions, feature extraction and classification recognition of motor imagery EEG signals. The system level performance are mainly decided by whether the system can quickly and efficiently extract the typical characteristic information from the signals. Using the method based on wavelet and wavelet packet decomposition, point at the C3 and C4 electrode channels, the component of EEG signals were outstanding to different scale to amplification study. To extract the characteristics of EEG signals include: change coefficient, volatility index and wavelet entropy.Select three kinds of classification algorithms: SVM, bayesian classifier, the BP neural network. Compared the classification time and the accuracy. Finally we concluded that the support vector machine is the best than others. Then, feedback the results and accuracy. The highest accuracy can reach 98.32%.