A method for identifying high-frequency steady-state visual evoked potential signals based on ensemble empirical mode decomposition

Abstract

In brain-computer interface (BCI) systems, the accurate identification of steady-state visual evoked potentials (SSVEP) is crucial in fields such as biomedicine, and various artifacts affect the identification accuracy making the signals difficult to identify. To identify steady-state visual evoked potentials, analysis is generally performed in the frequency domain to characterize the periodicity of the signal, more specifically, spectral analysis is required to characterize the distribution of electroencephalogram (EEG) signal power along the frequency. The high-frequency SSVEP has advantages such as less ocular stimulation for the subject but is difficult to identify. A method based on ensemble empirical modal decomposition (EEMD) is proposed to address the difficulty of identifying high-frequency steady-state visual evoked potentials. Decompose the three-channel data of Oz, O1 and O2 obtained from the experiment, perform spectrum analysis on the intrinsic mode function obtained by the decomposition, and select some effective intrinsic mode functions according to the obtained spectrogram and recombine them into new signals to replace the original signal. The multivariate simultaneous exponential (MSI) algorithm combined with the ensemble empirical modal decomposition was used to classify the reconstituted obtained signals, and the EEMD-MSI algorithm obtained better results compared with the direct classification method using the MSI algorithm.The average accuracy of the EEMD-MSI algorithm was improved by 9.957% and 16.424% compared with the MSI algorithm under the 3s and 4s time windows, respectively, and the highest accuracy reached 74.97%, 92.12%, respectively.