Research On Signal Processing Methods Of Brain-Computer Interface Basedon SSVEP

Brain-computer interface technology makes the information interaction between the brain and external devices become a reality.This field has broad application prospects and excellent research value,which brings good news for patients with motor dysfunc-tion caused by diseases or accidents.Steady-state visual evoked potential(SSVEP)has become an essential branch of the BCI system due to its advantages of high signal-to-noise ratio,less user training,and high information transmission rate.To improve the accuracy of the frequency recognition algorithm when there are many stimulus targets,and the length of the SSVEP signal is short,researchers have developed the signal processing algorithm of the SSVEP-BCI.Based on this,this paper improved the performance of the SSVEP-BCI by mining the useful information of SSVEP and studying from the angle of the frequency recognition algorithm.The research results are summarized as follows:(1)Based on the relevant literature,the stimulus source and stimulus frequency were selected.The stimulus source is the visual stimulus interface of LCD,and the gray value of the visual stimulus block is controlled by a sampled sinusoidal stimulation method to reduce the eye fatigue of subjects.The SSVEP signals of 12 subjects were successfully collected by the EEG acquisition module of OPENBCI.(2)In the traditional Canonical Correlation Analysis algorithm,local time informa-tion is not considered.The SSVEP signal is a non-stationary physiological signal which changes slowly with time.The time structure information is helpful for the analysis of physiological signals.This paper presents a Canonical Correlation Anal-ysis algorithm based on local time information(TCCA).The SSVEP data set of the brain-computer interface research group of Tsinghua University was used to optimize the local time range ?,and the comparative experiments of TCCA and CCA in dif-ferent time windows were carried out.The off-line experimental results show that the accuracy of TCCA under any time window is significantly better than CCA,and the average accuracy is increased by 2%~4%.When the time window is 1.5s,the average accuracy reaches 90.93%(40 stimulation targets).(3)Combining the TCCA algorithm with the filter bank CCA(FBCCA)algorithm,the FBTCCA algorithm is proposed,which considers harmonic information and time structure information together.In the off-line experiment,the EEG data of 6 subjects were used to optimize the parameters of the algorithm,and the EEG data of 9 subjects were used to compare FBTCCA and FBCCA in different time windows.The results show that the average accuracy of FBTCCA is slightly better than that of FBCCA,and the difference in algorithm accuracy in the short time window is more significant.When the time window is less than 0.8s,the average accuracy is increased by more than 1.6%.When the time window is 1.5s,the average accuracy is 94.6%(40 stimulation targets).This study confirms that TCCA and its derivative algorithm have potential in SSVEP-BCI in a short time window.(4)The multivariate empirical mode decomposition(MEMD)of SSVEP is carried out,and the influence of various intrinsic mode functions(IMF)components on the classification accuracy of SSVEP is discussed.It is found that part of the effective in-formation may be ignored when the IMF is screened to reconstruct its composition.In this paper,a signal purification algorithm based on grid search MEMD(GS-MEMD)is proposed.The weight of each IMF component in the reconstructed signal is deter-mined by grid search.GS-MEMD-CCA and GS-MEMD-MSI are proposed by com-bining them with CCA and MSI,respectively.Experimental results show that the new reconstructed signal improves the accuracy of the algorithm.And GS-MEMD-MSI al-gorithm has the highest average accuracy,which reaches 95.24%in the 2 second time window(6 stimulation targets).