Multiple Domain Feature Extraction Of Hand MI-EEG And The Design For Personalized Eeg Acquisition System

In recent years, the development of basic Brain Computer Interface(BCI) application and neurological research have accelerated the converge step of computer science, medicine, biology and other disciplines. New generation of BCI technology gradually combinedwith the more advanced technology and intelligent tools.The diversity of BCI applicationsslowly appeared. Brain research around the world has been risen to national strategies and important progresshas also been made in brain sciencein our country. BCI applications have been gradually engaged inliving entertainment, military, aviation area and other diversified direction. Many domestic and foreign research institutions have change their study object from animal to human being. Their exciting research results made a solid foundation for BCI using in human life.When people doing motor imaging,the activated brain regions is consistent with doing actual movement As a reslut, doing MI-EEG(Motor Imagery Electroencephalograph,) research isvery significantand meaningful.In this paper, two classes of MI-EEG are analysed by feature extraction, multi-domain feature combination, besides the correlation coefficient of between each feature are also computed to improve the recognition rate of MI-EEG.In order to promote the application of MI-EEG in the field of rehabilitation and enhance the strength of the EEG signals, personalized MI-EEG training system is designed in this paper for the purpose of improving the quality of brain electrical signal from where it appearing. Main research contentin this paperisas follows:(1) Feature extraction based on non-linear analysis and OEMDIn this paper, the nonlinear features of MI-EEG are extracted by Lyapunov exponent, Approximate Entropy(Ap En) and Sample Entropy(Samp En). These nonlinear feature confirmed the chaos characteristic EEG. Besides, the mean energy of intrinsic mode is also extracted uing orthogonal empirical mode decomposition(OEMD). After computing the linear correlation coefficient of Samp En and mean energy of intrinsic mode, it can be conclude that simply combining different features is not helpful for getting high classification rate, because the strong relationship may exist between them and do not have complementary characteristics. Correlation analyses between different features is an important method in feature combination.(2) Multi scale phase synchronization based on OEMD and feature combination.In order to get high recognition rate of MI-EEG,multiscale phase synchronizationindex(m PSI) is put forward as afeature.extraction method.Feature m PSI of each IMF component for brain signalsis computed from the process of OHHT. Combining m PSI, OHHT energy feature and common spatial subspace decomposition(CSSD) together as the multiple features of two class ECo Gfrom BCI competition III provided by Graz University, the best accuracy of it,is 96% using incremental support vector machine. The linear correlation within three types of features has shown the effectiveness of EEG phase feature and the complementarotycharacteristics of them。(3) Individual MI-EEG acquisition system design based on mirror virtual.The recognition accuracy of different classes MI-EEG is not only affected by feature extraction method and classifier but also depend on the quality of original EEG. In order to enhance the strength of EEG when subjects doing imagery task and activate the inner desire of the subjects, mirror therapy in the field of rehabilitation is used in MI-EEG training and acquisition process. A new personalized MI-EEG acquisition system basedon mirror virtualis designed in the Qt Creator development environment to improve the quality of EEG..In the EEG training mode, personalized visual and auditory stimulation are used with data gloves. Trainners wear the data gloves and do open and close hand activities, the action data of the hand are transmit to computer through serial port and can be used to control the hand animations in the interface. In training mode, trainers observe mirrored hand animation corresponding to the real action of the other hand which can enhance mirror neuron electrical activity in their brain and enhace the quality of MI-EEG.