Research And Implementation Of MI-BCI System Based On ERN Feedback

Brain computer interface includes many patterns,including motor imagery Brain-Computer Interface(MI-BCI),which decode changes in EEG rhythm through motor imagery,autonomously controls external devices and has a wide range of applications.In this paper,in order to control the advance,left turn and right turn of intelligent robot with EEG,the single detection error related negative potential(ERN)is used as the breakthrough point to enter the motor imagery classification for autonomous recognition errors of the system,and focus on the problem of the poor feedback effect in the MI-BCI engineering system.It is mainly to solve a series of engineering technical problems,such as the structural design of MI-BCI system based on ERN feedback,the design of experimental paradigm,the extraction of motor imagery features,the ERN feedback device technology,the ERN single feature extraction and detection technology,and the establishment of a system control platform.The main completion work is as follows:(1)The ERN experimental paradigm induced by online motion motor imagery.The experiment is based on a motor imagery brain-computer interface on-line system,which uses the screen to display motor imagery recognition results to induce ERN;and it locates the starting point of ERN by marking the display time of the screen,and completes the accurate positioning of the subject's perceived error time,providing a reliable time reference point for subsequent ERN time domain analysis.(2)The online processing system of motor imagery is designed.The core part of the MI-BCI system based on ERN feedback is the online classification and recognition of the EEG signals of motor imagery.The experiment compares the commonly used EEG feature extraction algorithms: wavelet transform,wavelet packet transform,AR model power spectrum estimation and Common Spatial Pattern(CSP)algorithm.The results show that the classification of multi-class CSP extraction features achieves the best accuracy,and it is determined that multi-class CSP algorithm is used to extract the motor imagery features for classification.(3)The ERN feedback device is designed.The core of the MI-BCI system based on ERN feedback is the motion imaginary classification of the system that identifies the error autonomously.When individuals perceive errors,a negative trending waveform that is specifically related to the wrong reaction is recorded in the central area of the scalp,namely ERN.Based on the physiological mechanism of ERN generated by individuals,a ERN feedback device for MI-BCI system is designed,and a ERN single detection algorithm is implemented to detect ERN in real time,so as to realize the classification of system independent recognition errors.(4)An ERN single-detection algorithm was designed.The ERN signal is weak,the wave amplitude is about 10?V,the signal-to-noise ratio is low,individual differences are large,and waveforms are usually obtained by average.In order to solve the problem of difficulty in single extraction of features,the brain electrical pathway is first selected based on the pattern of brain activity induced by ERN;then the wavelet transform method is used to extract the time domain characteristics of ERN in the low frequency range and the time domain and frequency in the high frequency range.Finally,with this low-dimensional feature combination,a 70.4% ERN single-detection accuracy rate was obtained.(5)The MI-BCI system based on ERN feedback was implemented.64-lead EEG acquisition system adopts EEG signal;motor imagery online processing module is designed to classify and recognize motion imagination task;The ERN is evoked by the recognition result screen;The ERN single-detection algorithm is used to detect ERN in real time to identify the wrong motor imagery classification;the motor imagery after correction was classified and coded;WI-FI communication is used to send command codes to control the robots.For this system,a number of subjects were tested.The experimental results show that the MI-BCI system has achieved self-identification of the wrong motor imagery classification,which improves the highest and average control accuracy of the external devices is 17.3% and 6.1%.