The Recognition Algorithm Research On The Steady-state Visual Evoked Potential

Brain-computer interface(BCI)is a novel appoach for information communication and control.It can provide direct communication and control between the human brain and computers or other electronic devices.It does not depend on traditional peripheral nerve and muscle output channels.It is a hot research topic in brain science.The control signals in BCI includes P300,sensorimotor rhythms,steady-state visual evoked potential(SSVEP),motion-onset visual evoked potential,and the hybrid signals among them.Due to its high signal-to-noise ratio and short response time,SSVEPs become one of the most commonly used control signals which are applied to the design and develop BCI systems.The BCI system based on SSVEP has a very high information transfer rate(ITR)and less training time,making this type of BCI a hot topic in current BCI research.Effective frequency identification methods are critical to develop high-performance BCI systems.This thesis will mainly study the frequency identification algorithm.The main contents and results are summarized as follows:1)A new frequency recognition method based on temporally local information constraint for multivariate synchronization index(MSI)was proposed.This method introduces the Tukeys tricube weighting function in the modeling of the covariance matrix to effectively leverage the temporally local information of the signal.Evaluated on the real EEG data experiments,the results demonstrated that the proposed algorithm effectively improved the performance of the original MSI algorithm.2)A new frequency recognition algorithm based on combining canonical coefficients was proposed.For the multichannel EEG signals analysis,canonical correlation analysis(CCA)can yield multiple canonical coefficients.In the traditional frequency recognition based on CCA,only the largest coefficient is selected,and other coefficients are discarded,and useful information that can be used for frequency identification is lost to some extent.To this end,we proposed an novel algorithm that combines all the canonical coefficients for classification.The experimental results on the open benchmark dataset showed that the proposed method significantly improved the performance of the standard CCA based algorithm.3)An new algorithm based on the filter bank likelihood ratio test(LRT)was proposed.In the field of signal processing,filter bank technology is widely used to process and analyze signals containing multiple sub-band frequency components.By introducing this technology,the performance of the original algorithm can be significantly improved.In this thesis,for the first time,we applied filter bank technology to the algorithm based on likelihood ratio test to further enhance its performance.Through the evaluation on the open benchmark dataset,the filter bank LRT method significantly enhanced the classification accuracy and ITR of the original LRT method.