Research And Implementation Of Online Brain-Computer Interface System Based On EEG Rhythm

Brain-Computer Interface(Brain-Computer Interface,BCI)is a new type of humancomputer interaction technology based on electroencephalogram(EEG)signals.By recording and decoding the regular features of the EEG,it establishes a direct information communication channel between the brain and the external devices that does not depend on peripheral nerves and muscles.This technology allows those people with disabilities whose nerves and muscles are damaged but whose brains still function normally to communicate with computers or external devices.The brain-computer interface based on motor imagery uses the change in mu / beta rhythm of the sensorimotor cortex of the brain as the detection basis,thereby continuously being the current research hotspot in the area of BCI technology.In recent years,various theories and algorithms on Motor Imagery BCI(MIBCI)have been proposed one after another to improve its performance to a certain extent.However,there are still many problems in the process of MIBCI system becoming practical,such as low recognition accuracy and difficulty in asynchronous control.As the core technology of the BCI system,the feature extraction and classification algorithms will directly affect the performance of the entire BCI system.In this thesis,first,we studied the selection of BCI parameter based on Common Spatial Pattern(CSP)through offline analyses,then three classification methods based on the features of Motor Imagery EEG(MIEEG)are compared,finally,the performance of CSP and Independent Component Analysis(ICA)are compared from the perspective of self-testing and migration testing.Because of the simple design and high recognition rate of the CSP spatial filter and zero training classifier,this thesis uses the One Versus One Common Spatial Pattern(OVO-CSP)and zero training classifier to build an online MIBCI system on the VC ++ platform.At the same time,the alpha rhythm is used as a switch between motion imaging and idle state,effectively realizing asynchronous control of the BCI system.The main contents of this thesis are as follows:(1)The parameter optimization problem based on CSP is discussed and the performances of the three classification methods are compared.Parameters such as the frequency band of feature extraction,the number of CSP training samples,and the segment of motor imagery data were optimized through offline analyses.Based on the offline MIEEG and CSP feature extraction algorithms,the performance of the three classifiers is discussed.The results show that the zero training classifier has a higher recognition rate.In the classifier migration test,the highest recognition rate is obtained by combining the machine learning classifier and CSP for migration.(2)The performance of CSP and ICA are comparatively analyzed in terms of self-test and migration test.Experimental results show that CSP is higher than ICA in self-test results,and ICA is significantly better than CSP in migration test results.However,considering that the ICA design process is more complex and sensitive to lead distribution,while the CSP design process is simple and has good maneuverability and practicality,this thesis chooses CSP as the feature extraction algorithm of the online MIBCI system.(3)An asynchronous online BCI system based on EEG rhythm is realized.Aiming at the recognition rate and control method of the online BCI system,this thesis proposes to use OVOCSP to extract the features of MIEEG and combine them with the alpha rhythm to realize the asynchronous control,and then build a simple and practical online MIBCI system.The system has a simple human-machine interaction interface,which can display EEG data recorded online in real time,and provides real-time task classification and visual feedback.The subject controlled the start of motor imagery through the alpha rhythm generated when their eyes closed,and at the same time controlled the movement direction of the ball through the sensorimotor rhythm.(4)An experimental paradigm of the online MIBCI system is designed.The CSP filter model can be updated online to avoid the import and export of CSP models and parameters,which improves the performance and practicability of the system.Four subjects participated in the online experiment.The experimental results show that all the subjects can control the ball to move to the end,and the average recognition rate is 83.5%,with the highest recognition rate in a single experiment occasionally reaching 100%.The BCI system in this thesis is developed on the VC ++ platform with the characteristics of good portability,fast running speed and low algorithm complexity.