Background Accurate classification of MI from EEG signals is crucial for non-invasive BCIs, especially for individuals with motor impairments. However, existing methods often overlook the synergies across multiple frequency bands, limiting their discriminative power. To address this limitation, we propose DAS-LSTM, a hybrid framework that integrates FBCSP for multi-band feature extraction, a simplified LSTM variant with reduced gating complexity, and a dual attention mechanism that prioritizes task-relevant temporal and spectral features.
Accur...
