EEG-Based Eye State Recognition Using VMD-CWT and DCAM
TL;DR
This study proposes a novel method combining cascaded variational mode decomposition (VMD-CWT) and a dynamic context-aware mask (DCAM), which reduces the number of parameters and computational complexity, demonstrating its effectiveness and efficiency for EEG-based eye state recognition.
Abstract
Electroencephalography (EEG)-based eye state recognition often suffers from low accuracy due to noise interference in the collected signals. To address this issue, this study proposes a novel method combining cascaded variational mode decomposition (VMD-CWT) and a dynamic context-aware mask (DCAM). First, variational mode decomposition is applied to decompose EEG signals, and the Pearson correlation coefficient is used to identify and extract sensitive feature components. Subsequently, a multi-band adaptive weighting strategy based on continuous wavelet transform (CWT), together with feature cascading, is employed to generate time–frequency feature maps, achieving both noise suppression and precise extraction of key features. Furthermore, a frequency-adaptive pooling granularity mechanism is integrated into the traditional context-aware mask to construct the DCAM, enabling an adaptive attention mechanism tailored to the target EEG features. Finally, a DCAM network is built with DCAM as its core. Experimental results indicate that the proposed method achieves an accuracy of 95.67% on the EEG Eye State dataset, outperforming the best-performing baseline method by 8.34 percentage points. In addition, the proposed framework reduces the number of parameters and computational complexity, demonstrating its effectiveness and efficiency for EEG-based eye state recognition.