Study Electroencephalogram Classification Based On Motor Imagery

The Brain Computer Interface(BCI)research has drawn attention of scientists in brain science,rehabilitation engineering,biomedical engineering and intelligent information processing.The BCI system aims at creating new direct information interaction and communication channels between brain and computer,rather than depending on brain's normal output channels of peripheral nerves and muscles.This paper mainly studies about brain-computer interface system based on motor imagery signal.Based on the synchronization and desynchronization feature of motor imagery signal,the signal preprocessing,feature extraction and classification are studied and then a real-time brain-computer interface system is designed.An algorithm combining wavelet packet decomposition and the common spatial pattern is proposed for feature extraction.The principle of the WPD-CSP algorithm is as following: firstly the wavelet packet transform is used in the original signal of each channel,then selecting the wavelet coefficients associated with u rhythm and beta rhythm to replace the raw data of each channel,finally directly using CSP algorithm in the wavelet coefficients domain for feature extraction.The algorithm makes CSP algorithm projection matrix has a better adaptability since it considers the frequency domain characteristics of EEG signals.Finally based on the "one to rest" and "one to one" support vector machine classifier structure,a new two-layer support vector machine classifier is designed.Compared with traditional methods,the proposed classifier has better results.Finally,a real-time BCI experimental system is devolved,a large number of experiments were implemented to invettigate the performsnce of the system.The results show that,EEG can be collected and stored.,the features can be classified online by this system.