Surface EMG-Based Inter-Session Gesture Recognition Enhanced by Deep Domain Adaptation.

Area of Science:

• Biomedical Engineering
• Neuroscience
• Human-Computer Interaction

Background:

• High-density surface electromyography (HD-sEMG) records muscle electrical activity using 2D electrode arrays.
• HD-sEMG allows temporal and spatial analysis for next-generation muscle-computer interfaces (MCIs).
• Current sEMG gesture recognition primarily focuses on intra-session analysis, lacking standardized benchmarks for real-world MCI applications.

Purpose of the Study:

• To establish a benchmark database for HD-sEMG recordings of hand gestures.
• To propose a deep learning-based domain adaptation framework for enhanced inter-session gesture recognition.
• To evaluate the proposed framework against state-of-the-art methods on multiple datasets.

Main Methods:

• Collected HD-sEMG data from 23 participants performing hand gestures using an 8x16 electrode array.
• Developed a deep learning domain adaptation framework to address inter-session variability.
• Validated the approach on NinaPro, CSL-HDEMG, and a newly created CapgMyo dataset.

Main Results:

• The proposed framework demonstrated superior performance in intra-session gesture recognition.
• Significant improvements were achieved in inter-session gesture recognition compared to existing methods.
• The benchmark database facilitates reproducible research in HD-sEMG-based MCIs.

Conclusions:

• The developed HD-sEMG database and deep learning framework advance the field of muscle-computer interfaces.
• The approach effectively enhances inter-session gesture recognition, paving the way for more robust real-world applications.
• This work provides a foundation for future research in personalized and adaptive MCIs.