Attention-Enhanced sEMG-EEG Fusion Network for Continuous Joint Angle Estimation in Home-Based Rehabilitation

Despite the widespread interest in the fusion of electromyography (EMG) and electrophysiological (EEG) signals for upper limb motion prediction, research targeting continuous motion posture prediction remains limited. A novel multimodal method that utilizes a deep learning network to fuse EEG and EMG signals for continuous joint angle prediction is proposed. In this paper, a feature extraction method is designed, where spatial-temporal features from EEG and sEMG signals are extracted using Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU), respectively. These extracted features are fused through an attention mechanism to predict continuous elbow joint angles. Experimental results show that the proposed model achieves a mean absolute error (MAE) of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$4.2370^\circ$</tex>, outperforming the EMG-only baseline <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$(7.2521^\circ)$</tex> by 41.6%. These results confirm the effectiveness of EEG as a complementary signal in continuous regression tasks, enabling more accurate and robust motion intention decoding for home-based rehabilitation.