A novel energy-motion model for continuous sEMG decoding: from muscle energy to motor pattern

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Surface electromyography (sEMG) based gesture recognition typically requires extensive training data, limiting its practical application.
• Existing methods struggle to recognize a wide range of gestures without significant user-specific calibration.

Purpose of the Study:

• To develop a novel dynamic energy model for decoding continuous hand actions using limited sEMG data.
• To reduce the complexity of training for sEMG-based gesture recognition systems.

Main Methods:

• A dynamic energy model was proposed, considering kinetic and potential energy in fingers based on muscle activation and adaptive-coupling mechanisms.
• Hand movements were categorized into ten energy modes based on finger acceleration patterns.
• Independent component analysis and machine learning were employed to correlate sEMG signals with energy modes for gesture representation.

Main Results:

• The model successfully enabled participants to perform untrained hand gestures with 100% accuracy.
• The system demonstrated performance significantly above chance in decoding single-finger energy levels.
• Real-time control experiments showed a high success rate (over 95%) in tasks like controlled manipulation.

Conclusions:

• The proposed dynamic energy model effectively decodes continuous hand actions with speed and force information from minimal sEMG data.
• This approach significantly reduces the learning task complexity for sEMG-based gesture recognition.
• The findings suggest a more adaptable and efficient method for human-computer interaction using sEMG signals.