Transfer Learning Over Time and Position in Wearable Myoelectric Control Systems

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Human-Computer Interaction

Background:

• Wearable sensors facilitate myoelectric control systems for upper limb prosthetics and assistive devices.
• Variations in electromyograms (EMGs) due to time and electrode placement pose challenges for real-world myoelectric control.
• Limited initial training data restricts the adaptability of these systems.

Purpose of the Study:

• To evaluate the effectiveness of transfer learning in maintaining motion recognition accuracy for myoelectric control systems.
• To assess transfer learning's ability to compensate for long-term EMG variations and different electrode placements.
• To determine the practical efficacy of transfer learning in real-world myoelectric control applications.

Main Methods:

• Collected one-month-long EMG data from upper limb wearable sensors.
• Applied transfer learning algorithms to adapt to variations in EMG signals over time.
• Tested system performance across three distinct electrode placements.
• Quantified motion recognition accuracy before and after transfer learning application.

Main Results:

• Transfer learning significantly compensated for EMG variations across a one-month period.
• The algorithms successfully adapted to different electrode positions, enhancing practical usability.
• Accuracy recovery was limited when electrodes were placed in non-optimal "out-of-muscle" locations.

Conclusions:

• Transfer learning demonstrates practical efficacy in enhancing the robustness of myoelectric control systems.
• Findings suggest the need for further research into electrode configurations and long-term data recording strategies.
• Optimizing electrode placement is critical for maximizing the benefits of transfer learning in myoelectric control.