The Resrearch For The Improvemenet Method Of The SSVEP Based BCI

Brain-computer interface(BCI)provides a direct connection between the brain and the external equipment,the technology has great development prospect and research value.The technology could improve the qualities of the lives of the patients who have difficulties to get communication with the outside because of the serious disease and accident.The research topic of the paper is the improvement method of the steady-state visual evoked potential(SSVEP)based brain-computer interface.The advantages of this type of the BCI are obvious,such as the effectively generated typical signal,less training of the users and the higher information transfer rates.However,the type of the BCI also has some drawbacks need to be solved,such as the neglect of the characteric of the typical SSVEP signal,the arbitrary judgement of the target and so on.To improve the perfermance of the SSVEP based BCI systems,this paper investigated the whole process of the SSVEP based BCI system from the experimental paradigm to signal processing methods.The main results obtained in this paper are listed as bellow:(1)In the experimental design aspect,the sampled sinusoidal stimulation method is applied to generate different stimil frequencies according to the corresponding studies.The method mentioned above avoid the limitation of the refresh frequency and increase the number of the stimulis.Meanwhile,the differences between the stimulis is enhanced due to the phase in the modulation equation.(2)In the signal pre-processing aspect,the sinusoid-assisted multivariate extension of the empirical mode decomposition(SA-MEMD)is used in the SSVEP signal preprocessing stage,the method improve the signal noise radio of the SSVEP signal.Meanwhile,the canonical correlation analysis(CCA)method is used to classify the reconstructed SSVEP signal.The result shows that the classification accuracy could be improved,the best average 9 class accuracy are 90.9%(4 s).Compared to the standard CCA algorithm,the average accuracy could be improved around 2% at each length of the time window.(3)In the feature selection and classification aspect,the paper investigated the influences on the standard CCA algorithm caused by the characteric of the typical SSVEP,the results could be concluded as: 1).the initial time of the EEG signal has great effects on the standard CCA algorithm especially in the short time window length condition,however if the initial time exceeds 210 ms,the classification accuracy of the CCA algorithm could attain a stable value.2).the fundamental harmonic and the first harmonic of the typical SSVEP signal has high energy and the energy of the first harmonic is less than the fundamental harmonic and this character could not be reflected in the standard CCA algorithm.Therefore,the paper proposed the SSVEP character based SA-MEMD multi-subbands CCA algorithm,the algorithm analysis IMFs containing the first harmonics of the SSVEP signal,the experiment results show that the proposed algorithm could imperove the classification accuracy,the best average 9 class accuracy are 91.4%(4s).Compared to the standard CCA algorithm,the average accuracy could be improved around 3.5% at each length of the time window.(4)In the system processing structure aspect,the paper proposed an filter module in the whole processing structure of the system,the module could filter the fuzzy correlation value sequences and improve the classification accuracy.The filter module could make the BCI system more practical and more secure.The module filtered the correlation value sequences generated by the standard CCA algorithm and the SSVEP character based SA-MEMD multi-subbands CCA algorithm respectively.The results indicates that the average accuracy are 89.5%(4s)that is correspond to the standard CCA algorithm and 93.0%(4s)that is correspond to the SSVEP character based SA-MEMD multi-subbands CCA algorithm.When the SSVEP signal could be aroused effectively,the loss ratio is relative low which indicates that the instruction filter module could optimize the signal processing structure of the SSVEP based BCI system.