Limb-position robust classification of myoelectric signals for prosthesis control using sparse representations.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Non-invasive myoelectric prosthesis control aims to interpret user intentions from surface electromyography (sEMG) signals.
• Linear Discriminant Analysis (LDA) is widely used for real-time movement classification but struggles with performance degradation in varied conditions.
• Existing methods limit the practical utility of myoelectric prostheses due to sensitivity to environmental and training condition changes.

Purpose of the Study:

• To develop an enhanced classification method for myoelectric prosthesis control that is robust to deviations from training conditions.
• To improve the accuracy and reliability of recognizing intended upper-limb movements from sEMG signals.
• To demonstrate superior performance over LDA in recognizing untrained movements.

Main Methods:

• Proposed a novel classification approach utilizing sparse representations of sEMG time-frequency features.
• Constructed a data dictionary from sEMG time-frequency features for robust pattern recognition.
• Applied and evaluated the method in the context of upper-limb position changes.

Main Results:

• The enhanced method demonstrated significant accuracy improvements in untrained positions (7.95% increase, p ≪ .001) compared to LDA.
• Achieved substantial accuracy gains across all multi-position training conditions (p < .001).
• Showcased enhanced pattern recognition robustness against generic deviations from training conditions.

Conclusions:

• The proposed sparse representation method offers a more robust and accurate approach to myoelectric prosthesis control.
• This advancement can enhance the real-world utility and performance of prosthetic devices.
• The findings suggest a promising direction for improving human-machine interfaces in prosthetics.