Decoding SSVEP patterns from EEG via multivariate variational mode decomposition-informed canonical correlation analysis

Steady-state visual evoked potential (SSVEP) is one of the most popular neural patterns used to develop braincomputer interface (BCI). To address the issue of electroencephalogram (EEG) is easily interfered with noise artifacts and differences in SSVEP components between different channels and different frequency bands. We propose multivariate variational mode decomposition-informed canonical correlation analysis (MVMD-CCA) to improve the decoding performance of SSVEP patterns. Firstly, the multivariate variational mode decomposition method is used to decompose electroencephalogram into inherent mode functions (IMFs) of different frequency bands for minimizing the effects of artifacts. Then, the obtained inherent mode functions components of various frequencies and channels are weighted to reconstruct the electroencephalogram signal. Furthermore, it has a fast convergence speed and good stability when using the sparrow search algorithm (SSA) to optimize the weight parameters. Finally, the weighted reconstructed signal is classified by the canonical correlation analysis (CCA). An extensive experimental analysis is implemented with electroencephalogram collected from nine subjects use an eight-target SSVEP system. Results show that the multivariate variational mode decomposition-informed canonical correlation analysis significantly outperforms the canonical correlation analysis, with a maximum increase of 14.2% in SSVEP decoding accuracy. Simultaneously, the information transfer rate (ITR) increased by 6.5. An extensive comparison was performed between the proposed method and other competing approaches, including multivariate synchronization index (MSI), temporally local multivariate synchronization index (TMSI), and filter bank canonical correlation analysis (FBCCA). The superiority of our method demonstrates its great promise in the development of improved BCI systems.