Research On Algorithms Of Left Vs. Right Hand Motor Imagery Based BCI

Brain-Computer Interface (BCI) is a novel human-computer interface. It is a system that transfers brain information and realizes control by using the computer or other electrical devices to analyze the brain activities under specific environment and conditions, without depending on brain's normal output channels of peripheral nerves and muscles. Thus, the research on BCI is significant to rehabilitation engineering and the development of new communication and control approaches. It has been drawing more and more attention of scientists from the fields of brain cognition research, rehabilitation engineering, biomedical engineering, automatic human-machine control, and so on.Feature extraction and pattern classification of EEG is a pop problem during the research of BCI. Oriented to the left vs. right hand motor imagery based BCI, offline research on the algorithms of feature extraction and pattern classification was carried out for the small number of samples from the public data sets of 2003 and 2004 International BCI Algorithm Contests. At the beginning, feature bands were selected by power spectra analysis of the EEGs from channel C3 and C4 based on the properties of ERD/ERS. Then, non-parameter features were extracted based on the AR model and the Morlet wavelet transform. Fisher linear determinant classifier was used to compare the different selection of feature bands, time windows and feature extraction methods. Finally, classifiers designed by Fisher LDA, SVM and Bayes theory were studied with the features based on Morlet wavelet transform.The experimental research on the left vs. right hand motor imagery based BCI was implemented based on the former research. First of all, an experimental system, as well as an experimental plan, was designed with the 128-channel EEG analyzer for the left vs. right hand motor imagery based BCI. EEGs from 2 subjects were recorded successfully. Classifiers were attained by the training set after feature bands selection and feature extraction, and tested by the testing set, which validated the reliability and transplantability of the algorithms. Useful experience was gained for the on-line real time BCI experimental research.