The Research Of Motor Imagery Electroencephalogram Processing And Classification

Electroencephalogram(EEG)is a kind of physiological electrical signal,which is produced by the numerous nerve organizations collaboratively when they are conducting the electrophysiology activities in the brain cortex.Scientific researches show that analyzing the EEG can understand humans' thought and consciousness awareness.Brain Computer Interface(BCI)is one of the important application directions on the EEG,and it is a man-machine interactive system that does not depend on the brain's normal output pathways of peripheral nerves and muscles.Currently,the researches about the two kinds of signal recognitions based on the motor imagery of left and right hands have been relatively mature.However,there exists the problems of low recognition rate,poor real-time performance,and susceptibility to disturbance in the researches about the multi-dimension motor imagery EEG.Therefore,the research about the multi-dimension motor imagery EEG has been a hot focus,and how to process and recognize the multi-dimension motor imagery EEG is a challenge.The thesis develops a deep research on the process and recognition of the four kinds of motor imagery EEG from the left hand,right hand,right leg and tongue,the signal used is EEG of C3?C4?Cz and CP4.Based on the experimental condition monitoring need,O1/O2 2 leads are added to conduct the state transition.Moreover,the control of making the vehicle move forward,move backward,turn left,turn right,as well as start and stop in the virtual scenario has also been achieved.And the thesis' s specific work and innovation include:(1)The de-noising algorithm is researched since the acquisition of EEG contains many other disturbing signals,such as electrooculogram(EOG),electromyography(EMG),electrocardiogram(ECG),power noise and et al.And a new threshold de-noising algorithm based on dual-tree complex wavelet transform(DTCWT)is proposed.The simulation results show the advantages of the proposed algorithm.(2)Existing researches reveal the phenomenon that there is inconsistent EEG intensity in these two scenarios: one is that different testers are asked to perform the same motor imagery and the other is that the same tester is asked to perform the same motor imagery.As to this problem,the normalized enegy feature of alpha and beta brain-wave is proposed,and simulation results show that using normalized waveletpower spectrum based on alpha and beta brain-wave can obtain better performance than using the power spectrum that has not been normalized.Moreover,the sample entropy feature based on alpha and beta brain-wave is proposed on the ERS/ERD phenomenon.The power spectrum and the improved sample entropy are combined to be form a new feature,which can not only improve the correctness rates of classification significantly,but also decrease the calculation complexity.(3)As to the problem that there exists low calculation efficiency and recognition rate in the traditional pattern recognition algorithm for the motor imagery EEG,the cross-validation and LOO error correction are utilized to optimize least square support vector machine classification algorithm,which can not only improve the correctness rates of classification significantly,but also decrease the calculation complexity.On the condition of using the mixed feature,the average recognition rate can achieve70.96%,and compared with the grid-support vector machine algorithm,the average calculation time can also achieve a 0.45 s decrease.(4)BCI online platform controlled virtually is designed to verify classification results of the motor imagery EEG.The four kinds of motor imagery EEG from the left hand,right hand,right leg,tongue are processed and recognized,then the results from the recognition are transformed into the commands,which are transmitted to the virtual reality scenario,to achieve the control on these four virtual actions,that is,moving the vehicle move forward,move backward,turn left,turn right,as well as start and stop.The opened-eye or closed-eye state of O1/O2 channel is utilized to conduct the switching of the working or sleeping state of the experiment.The recognition rate of synchronous online experiment results can reach 67%,that of asynchronous online experiment based on four kinds of motor imagery EEG are improve in some degree.