Robust decoding of hand movement for mitigating muscle contraction variability from surface EMG signals

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Surface electromyography (sEMG) pattern recognition is crucial for intuitive control of artificial limbs.
• Muscle contraction variability in sEMG signals degrades the accuracy of hand movement decoding.
• Existing decoding systems struggle with performance inconsistencies due to varying muscle activation levels.

Purpose of the Study:

• To develop a novel framework for robust hand movement decoding from sEMG signals.
• To mitigate the impact of muscle contraction variability on decoding accuracy.
• To enhance the reliability of sEMG-based control systems for prosthetic devices.

Main Methods:

• Collected two-channel sEMG data from eight subjects performing six hand movements across three contraction levels.
• Extracted 15 sEMG features and employed five classic classifiers for decoding.
• Proposed a simulation-driven feature mapping framework using a Feature Denoising Network to normalize sEMG features.

Main Results:

• Decoding accuracy decreased by over 32% without the framework under varying contraction levels.
• The proposed framework significantly improved decoding accuracy, with gains of at least 26.99%.
• The Feature Denoising Network effectively mapped abnormal sEMG features to normal levels, restoring performance.

Conclusions:

• The developed framework offers a feasible solution to the challenge of muscle contraction variability in sEMG decoding.
• This approach enhances the robustness and accuracy of hand movement intention recognition.
• The findings have significant implications for the development of advanced myoelectric hand prostheses.