Domain Adaptation With Self-Guided Adaptive Sampling Strategy: Feature Alignment for Cross-User Myoelectric Pattern...

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Surface electromyographic (sEMG) signals are crucial for gestural interfaces.
• Individual differences in sEMG signals pose a significant challenge for cross-user control systems.
• Existing unsupervised domain adaptation (UDA) methods struggle with instantaneous data distribution during model updates, especially with large domain shifts.

Purpose of the Study:

• To develop a novel UDA method addressing the cross-user variability in sEMG signals.
• To improve the feature representation alignment of myoelectric patterns across different users.
• To enhance the robustness and accuracy of myoelectric pattern recognition for user-independent control.

Main Methods:

• Proposed a UDA model integrated with a self-guided adaptive sampling (SGAS) strategy.
• Utilized domain distance in a kernel space to select reliable samples for classifier updates.
• Recorded sEMG data from nine subjects performing six finger and wrist gestures.

Main Results:

• The proposed UDA with SGAS achieved a mean cross-user classification accuracy of 90.41% ± 14.44%.
• Demonstrated statistically significant improvement over state-of-the-art methods.
• Showcased enhanced alignment of myoelectric pattern features across users.

Conclusions:

• The novel UDA framework with SGAS effectively addresses cross-user variability in sEMG.
• This method offers a promising tool for developing multi-user and user-independent myoelectric control.
• The SGAS strategy improves feature representation by screening reliable instantaneous samples.