Related Experiment Video
Updated: Jan 9, 2026
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
1.1K
MoEMba: A Mamba-based Mixture of Experts for High-Density EMG-based Hand Gesture Recognition
Summary
The MoEMba framework improves high-density surface electromyography (HD-sEMG) gesture recognition accuracy. It addresses inter-session variability, enhancing human-computer interaction (HCI) systems.
Area of Science:
- Biomedical Engineering
- Human-Computer Interaction
- Machine Learning
Background:
- High-density Surface Electromyography (HD-sEMG) is crucial for Human-Computer Interaction (HCI), providing muscle activity data.
- Low inter-session and inter-subject classification accuracy hinders practical HD-sEMG applications due to signal variability.
Purpose of the Study:
- To introduce the MoEMba framework to enhance HD-sEMG-based gesture recognition.
- To improve robustness against session-to-session variability in HD-sEMG signals.
Main Methods:
- Developed the MoEMba framework utilizing Selective State-Space Models (SSMs).
- Incorporated channel attention for temporal dependencies and cross-channel interactions.
- Integrated wavelet feature modulation for multi-scale temporal and spatial signal representation.
Main Results:
- MoEMba achieved a balanced accuracy of 56.9% on the CapgMyo HD-sEMG dataset.
- Demonstrated superior performance compared to state-of-the-art methods.
- Showcased robustness to session-to-session variability and efficient handling of high-dimensional data.
Conclusions:
- The MoEMba framework significantly advances HD-sEMG-based gesture recognition.
- The approach offers a robust solution for variability challenges in HD-sEMG signals.
- MoEMba holds potential for improving future HD-sEMG-powered HCI systems.
