CSAC-Net: Fast Adaptive sEMG Recognition through Attention Convolution Network and Model-Agnostic Meta-Learning.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Surface electromyography (sEMG) is crucial for bionic limb control but suffers from subject specificity and electrode placement variability.
• Developing adaptive models for new subjects remains a significant challenge in sEMG-based gesture recognition.

Purpose of the Study:

• To introduce CSAC-Net, a novel deep neural network designed for robust and adaptive sEMG gesture recognition.
• To improve the adaptability of gesture recognition models for new subjects in the context of bionic limb control.

Main Methods:

• Extraction of time-frequency features from raw sEMG signals.
• Implementation of a convolutional neural network (CNN) with an attention mechanism for advanced feature extraction.
• Application of model-agnostic meta-learning (MAML) for rapid adaptation to new subjects.

Main Results:

• CSAC-Net demonstrated superior performance compared to state-of-the-art methods in adapting to new subjects.
• Ablation studies confirmed the effectiveness of the proposed network architecture and learning strategy.
• Experiments on the CapgMyo dataset validated the advancements of CSAC-Net.

Conclusions:

• CSAC-Net offers a significant advancement in adaptive sEMG-based gesture recognition for bionic limb control.
• The integration of time-frequency features, attention mechanisms, and meta-learning provides a powerful approach for personalized human-computer interaction.