Deep Residual Network With Channel Attention for Improving Hand Gesture Recognition With Surface Electromyography Signal

Hand gesture recognition based on surface electromyography (sEMG) is a key enabling technology for human-computer interaction and intelligent prosthetic devices. Despite its importance, consistent and accurate recognition is difficult to achieve. This difficulty arises from the non-stationary nature of sEMG signals, their intricate temporal dynamics, and substantial differences across users. To address these challenges, this study introduces a robust deep learning architecture named CNN-ResBiGRU-SE, designed to improve gesture classification robustness and accuracy. The proposed framework comprises three principal modules. First, convolutional neural networks are employed to extract discriminative spatial features from multi-channel sEMG recordings. Second, a residual bidirectional gated recurrent unit is incorporated to capture long-range temporal dependencies while stabilizing learning. Third, a squeeze-and-excitation attention mechanism dynamically reweights feature channels, enabling the model to focus on the most informative signal components. Extensive evaluations are performed using the publicly available NinaPro-DB1 and NinaPro-DB5 datasets. These experiments cover a wide range of gesture types and inter-subject variations, thereby providing a comprehensive assessment of the model's generalization capability. The CNN-ResBiGRU-SE framework attains recognition accuracies of 89.14% on NinaPro-DB1 and 92.59% on NinaPro-DB5. These results surpass those reported by existing state-of-the-art approaches. The findings confirm that integrating residual temporal modeling with channel-wise attention yields a reliable and effective solution. This combined strategy demonstrates strong potential for improving the performance of sEMG-based hand gesture recognition systems in practical applications.