Feature Extraction And Classification Of EEG For Imagery Movement

Brain-Computer Interface (BCI) is a new way of man-machine interface. It does not pass through the brain's normal output channels of peripheral nerves and muscles, but directly obtains information from brain to communicate with the outside world. One of the most important purposes for the BCI comes mainly from the hope that the system can provide language communication and environmental control for those with severe motor disabilities but normal thoughts, and improve the quality of their lives. What's more, BCI technology also has potential applications in other fields, such as special works, military affairs and entertainment. During these years, as a multidisciplinary cross technology, BCI research has drawn attention of scientists in brain-science research, rehabilitation engineering, and biomedical engineering or human machine automatic control. The theory research such as feature extraction, classification and experiment research of electroencephalogram(EEG) play important roles.Under the condition of no stimulation during the mental tasks, mental EEG can be obtained direcetly through left-right hands imagery movement. Therefore, the thesis analyzes theory methods such as feature extraction, classification and signal acquisition, experiment design of BCI system for imagery movement EEG. The main contributions of the paper are as follows:(1) A EEG recognition algorithm based on discrete wavelet transform and BP neural networkIn BCI of imagery movement, a new EEG recognition method (DWT-BP method) which combines discrete wavelet transform (DWT) with BP neural network is proposed to solve the problems such as the low classification accuracy and weak anti-disturbances etc. In DWT-BP, a rational time window is set through calculating the average power of C3 electrode and C4 electrode in left-right hands imagery movement. Then the average power during the time window is taken into DWT. The combinational signal of approximation coefficient A6 on the sixth level is selected as a signal feature and BP neural network is used as classifier to analyze the observed EEG data. The experiment results on"BCI Competition 2003"competition database show that the proposed method could accurately extract substantial features of EEG and display the better anti-disturbances and classification performance. The recognition rate reaches 94%. It proves that the method is effective for EEG recognition of imagery movement, and makes a foundation for realizing on-line BCI systems of imagery movement.(2) Neural network ensemble used for EEG classificationBP neural network is able to approach arbitrary non-linear function, and it has strong self-learning, adaptive capability. Therefore, as a EEG classifier, BP neural network has widely application in BCI. However, for lack of prior knowledge, it is difficult to find suitable network structure used BP neural network algorithm, which affects generalization capability of classifier. A recognition algorithm based on neural network ensemble(NNE) is presented to solve this problem.The main features of left-right hands imagery movement are obtained by using DWT, then a NNE model is constructed by Bagging, and the output result of NNE is received through the relative majority vote. The experiment results on"BCI Competition 2003"competition database show that the recognition rate is better than that of the single neural network(99.3%). Moreover, the configuration difficulty of single neural network is reduced, and the generalization ability of the system is improved.(3) Related experiment design for imagery movement mental EEGBecause a majority of experiment data of BCI research at home comes from abroad, the thesis designs three different experiment program to acquire EEG. In the first experiment, the subject is asked to press the key using left hand or right hand by means of the direction of light. The subject is asked to imagine left or right hand movements into the direction of the arrow in the second experiment. In the final experiment, the visual stimulation of the first experiment is replaced by auditory stimulation. The subject is asked to close his eyes, and in accordance with"left"and"right"voice to press the key. Use the feature extraction and classification algorithm presented to recognise collected EEG at above three circumstances, all of which proves that the experiment program is feasible and EEG recognition algorithm is effective.A BCI simulation control system is designed based on the previous work. The system recognises obtained EEG with the presented DWT-BP method, then control a car moving. And finally it achieves to control the direction of a car moving by thought.The study results presented in this thesis may make a significant contribution to promote the classification performance and generalization ability, all of which establish a substantial theory and experiment foundation for realizing on-line BCI systems based on imagery movement mental EEG. In addition, the summarized experience both in theory and experiment has an important reference value in design and application of BCI system based on imagery movement.