Research On The Key Technology Of EEG-fNIRS Based Few Channel And Bimodel Brain Computor Interface

Brain computer interface(BCI)is a new information transmission device that is established between the brain and the external electronic equipment and not dependent on muscle tissue or peripheral nerve,it shows wide application value in the fields of rehabilitation,entertainment and military etc.At present,the BCI technology becomes more practical and wearable,so the BCI with a few channels obtains increasingly attention.The brain signal induced by motor imagery is a kind of spontaneous signal,so it can be used as the asynchronous BCI system.Because of the deficiency of the single mode BCI,the study of the dual mode BCI has become a trend.In this paper,we mainly study the few-channel feature extraction method and the bimodal fusion method of the motor imagery BCI with a few channels which based on the electroencephalograph(EEG)and the functional near-infrared spectroscopy(fNIRS),and the following findings were achieved:(1)In order to make full use of the limited information of a few channels,a feature extraction method based on time-frequency-space three domains was proposed.Firstly,in frequency domain,channels were reconstructed by the multivariate empirical mode decomposition(MEMD)method to obtain frequency bands only related to motor imagery,then in time domain the reconstructed channels were extended by phase space reconstruction(PSR)to obtain more channels,finally the more channels were extracted features by common space pattern(CSP)in space domain.Comparing with the MEMD+CSP method and PSR+CSP method,we found that our proposed method achieved an average accuracy of 81.3%,while the other two methods achieved an average accuracy rate of 67.2%,78.6%respectively.The above results show that the combination of the three domain method is more able to mine the information of the few channels.(2)In order to solve the problem that the MEMD method has some disadvantages such as the aliasing and the time consuming,a new algorithm is proposed,which is named sin-MEMD.In this method,the n original signals were decomposed by MEMD with a sinusoidal signal as assisted signal together,and the sinusoidal signal provided reference for decomposition of the original signal,so as to reduce the modal aliasing and reduce the screening time of the intrinsic mode function(IMF).MEMD and sin-MEMD were used to decompose the EEG data,and the results showed that sin-MEMD could effectively reduce the modal aliasing and improve the computational efficiency.(3)In order to give full play to the advantages of dual mode fusion,after feature extraction of the EEG and fNIRS data,the best features of two mode were selected by quadratic programming feature selection method and then were classified by support vector machine(SVM).The results showed that the average accuracy was 87.1%,which was increased by compared with only EEG mode,and increased by about 30%compared with only fNIRS mode.The dual model's results of most of the subjects were better than any single model's results.When one modal reflection is not obvious,the other mode can still provide high accuracy.The above results proved that the dual mode has the advantage of both the correct rate and the robustness.