Classification of Motion Imagination via Multivariate Variational Mode Decomposition and Second-Order Blind Identification Using Multi-Channel Data

BCI devices based on motor imagery classification can significantly improve the quality of life for paralyzed patients. However, with the advancement of medical devices, clinical equipment is evolving towards portable devices, which have far fewer channels than clinical equipment. Portable devices can collect both Electroencephalogram (EEG) and magnetoencephalograms (MEG). Most current techniques are not suitable for low-channel data or are only applicable to either EEG or MEG. To address these issues, this paper proposes a classification method based on multivariate variational mode decomposition (MVMD) combined with Second-Order Blind Identification (SOBI). First, MVMD decomposes the original multi-channel signals into various components, and then SOBI is used to denoise these components. Next, multiple features such as mean, and fractal dimension are extracted from the clean components. Finally, classifiers like SVM are used for classification. On EEG and MEG datasets, classification accuracies of 74.2% and 82.5% were achieved, respectively. Compared to other advanced methods, our approach achieved higher classification accuracy.