Research Of EEG Signal Processing Based On Motor Imagery

Electroencephalogram (EEG) is the overall reflection of brain nerve cellselectrophysiological activity on cerebral cortex of scalp, which contains a large number ofphysiological and disease information. Because there are correlation between EEG andconscious state of brain, it is possible to identify different conscious states by classification ofdifferent EEG models and then forms a communication control system which don’t depend onthe normal brain peripheral nerve and muscle output channels, which is so-calledBrain-Computer Interface (BCI). Motor imagery means imagination of limb motor withoutactual physical action. EEG evoked during motor imagery has the characteristics of event-relateddesynchronization (ERD) and event-related synchronization (ERS) by which we can judge themovement intention and control external device. So motor imagery EEG becomes the mostfrequently used feature signal of BCI.The thesis designs and realizes a set of real-time online BCI experiment system based onmotor imagery EEG. The system has several parts including the acquisition, preprocessing,feature extraction, pattern classification and experiment system design of motor imagery EEG.For signal acquisition, the method of choosing electrode and its settlement, the way of EEGchannel link, and in addition the acquisition experiment design were described. For signalpreprocessing, four different types of digital filters and their effects were studied and contrastedrespectively, through analyzing the ellipse filter had the best effect. For feature extraction, fastfourier transform, short-time fourier transform and power spectral density algorithms wereapplied in the research of this field, during deeply research the quality and effect of all the threealgorithms were analyzed and evaluated respectively for achieving better results of featureextraction. For pattern classification, three algorithms were studied in the application of patternclassification of motor imagery EEG, including the mahalanobis distance classifier, fisher lineardiscriminative classifier and support vector machine classifier, through classificationexperiments the advantages and disadvantages of various classification algorithms werecompared so as to provide basis for choosing the best classifier of real-time BCI systems. Finally,the real-time online BCI experiment system based on motor imagery EEG was established onLabVIEW platform, and furthermore three types of experiment modes had been successfullyachieved and thus the feasibility and effectiveness of the algorithms were further verified.