Understanding of Task-Specific and Subject-Specific Components in Surface EMG.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Human-Computer Interaction

Background:

• Surface electromyogram (sEMG) signals are crucial for human-machine interfaces.
• Current models face challenges in generalizing across individuals due to unique neuromuscular traits.
• This limits the effectiveness of sEMG in gesture recognition and user identification.

Purpose of the Study:

• To introduce a disentanglement model to separate task-specific and subject-specific components from sEMG signals.
• To enhance the generalization and interpretability of sEMG-based gesture recognition and user identification systems.
• To improve the robustness of sEMG applications in real-world scenarios.

Main Methods:

• Developed a disentanglement model to process sEMG signals.
• Separated sEMG signals into task-specific and subject-specific components.
• Evaluated the model's performance on gesture classification and user identification tasks across different subjects and days.

Main Results:

• Disentangled task-specific components significantly improved accuracy in both gesture classification and user identification.
• The model outperformed conventional methods in cross-subject and cross-day scenarios.
• Task-specific components captured consistent gesture patterns, while subject-specific components reflected individual neuromuscular characteristics.

Conclusions:

• The disentanglement approach enhances sEMG-based classification performance and interpretability.
• Extracted components offer insights into physiological mechanisms underlying sEMG signals.
• The model shows promise for improving real-world sEMG applications like rehabilitation and authentication.