Research on EEG signal decomposition and classification based on grey wolf optimized MVMD

Motor Imagery (MI) decoding in Brain-Computer Interface (BCI) is challenging due to the complex nature of Electroencephalography (EEG) signals, which exhibit strong non-stationarity, low signal-to-noise ratio, and high inter-subject variability. These factors hinder effective feature extraction essential for accurate classification. To overcome these issues, this study introduces an advanced EEG signal enhancement and reconstruction framework tailored for MI decoding. The approach consists of two key components: First, an improved Multivariate Variational Mode Decomposition (MVMD) method, optimized using Grey Wolf Optimization (GWO), which automatically selects the best decomposition parameters to achieve accurate Intrinsic Mode Function (IMF) separation. Second, a feature reconstruction strategy utilizing Pearson correlation coefficients and information entropy to compute modality-specific IMF weights, followed by a weighted summation to enhance EEG signals. Experimental results on the BCI Competition IV-I dataset validate the effectiveness of this framework, achieving an 86.20% classification accuracy— significantly outperforming traditional methods. This optimized approach offers a reliable and efficient solution for MI decoding in real-world BCI applications.