A Real-Time High-Density sEMG Gesture Recognition System Distilled from a Deep Model.

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Surface electromyography (sEMG) is crucial for human-machine interaction (HMI) gesture recognition.
• High-density EMG systems offer improved accuracy but face deployment challenges due to hardware and computational constraints.

Purpose of the Study:

• To develop a real-time, wearable high-density sEMG system for efficient gesture recognition.
• To balance performance and computational efficiency using knowledge distillation.

Main Methods:

• A 64-channel electrode array was used with a deep learning approach.
• A lightweight student model, distilled from a VGG-16 teacher model, processed sEMG signals.
• The system incorporated optimizations for online streaming, including sliding windows and robust calibration.

Main Results:

• Achieved an average recognition accuracy of 88.12% for 11 hand gestures.
• Maintained a processing time of 9.7 ms per segment, fulfilling real-time requirements.
• Demonstrated high usability in real-world scenarios with 12 participants.

Conclusions:

• The proposed system enables practical, wearable, high-density sEMG-based gesture recognition.
• Knowledge distillation offers an effective strategy for efficient deep learning deployment in sEMG applications.
• The system shows significant potential for real-time HMI applications.