Research On Online Asynchronous Brain-computer Interface System Based On Motor Imagery

Over the past decades,many regular features of electroencephalogram(EEG)has been found,one of them is energy changes caused by motor imagery.Therefore,brain-computer interface(BCI)was proposed to provide these users with communication channels that do not depend on peripheral nerves and muscles.The motor imagery EEG is initiative,and different modes of motor imagery can cause changes of specific EEG rhythms at different locations on the scalp,mainly presented as the changes of frequency band energy.Hence,the complete BCI system based on motor imagery was studied,that include EEG data acquisition,feature extraction after preprocessing,pattern classification,and generation of control commands.For asynchronous systems,idle state detection is also required to determine whether the user is in motor imagery state.As the kernel of a BCI system,algorithms of feature extraction and pattern classification directly affect the performance of the entire system.Therefore,feature extraction and classification algorithms were studied through off-line EEG data analysis firstly.The classification algorithm used in this paper is the support vector machine(SVM)which is based upon statistical theory.Based on the open-source EEG data and SVM classification algorithm,three frequency band energy feature extraction methods were discussed,i.e.,band-pass filtering,wavelet decomposition,and Hilbert-Huang transform.The changes of EEG band energy were verified.In the same frequency band,the classification accuracies of different energy extraction methods are very similar.Compared with the monopolar electrodes,the band energy characteristics of EEG data collected by the bipolar electrodes are classified better.In addition,the common spatial pattern(CSP)was studied as an EEG feature extraction algorithm.Due to band-pass filter is needed before CSP,it can be considered as a further feature extraction of the band energy.The results also verified that,under conditions of monopolar multi-electrode and multiple modes,the accuracies of the CSP algorithm were greatly improved compared to band energy algorithms.Since the EEG amplifier used in this project can acquire monopolarly EEG data with 16 channels,CSP algorithm has obvious advantages for these data.In order to increase the benefits of CSP algorithm,a frequency band selection method based on power spectral density(PSD)was proposed,which can select the more suitable frequency band according to the EEG characteristics of different subjects.In order to realize the online asynchronous BCI system,the idle state is regarded as a task and processed as the same as the motor imagery.So this system has three tasks.After off-line analysis of EEG data,practical electrodes,filter coefficients,CSP projection matrix and SVM classification model were obtained.An online BCI system was designed in this paper,which can classify EEG data through the models obtained by off-line analysis while acquiring.The commands of online system are represented by a demonstration software,in which the user can move the picture control to the target according to the interface,and the score was calculated.According to the results of online BCI system,although they are not good enough,the function of the online system was reflected effectively.