Related Experiment Video
Updated: Jun 27, 2025
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
458
Evaluation of Hand Action Classification Performance Using Machine Learning Based on Signals from Two sEMG
Hope O Shaw1, Kirstie M Devin1, Jinghua Tang1
1School of Engineering, Faculty of Engineering and Physical Sciences, University of Southampton, Southampton SO17 1BJ, UK.
Sensors (Basel, Switzerland)
|April 27, 2024
Summary
This study demonstrates high accuracy in myoelectric hand control using only two surface electromyography (sEMG) electrodes. Machine learning algorithms, particularly SVM, achieved performance comparable to multi-electrode systems, enhancing prosthetic functionality.
Area of Science:
- Biomedical Engineering
- Rehabilitation Technology
- Human-Computer Interaction
Background:
- Classification-based myoelectric control enables advanced prosthetic hand functionality.
- High accuracies often rely on multiple sEMG electrodes or additional sensors.
- Two-electrode sEMG systems are common but lack detailed performance studies.
Purpose of the Study:
- To investigate the classification performance of signal processing and machine learning algorithms using two sEMG electrodes.
- To compare the effectiveness of LDA, KNN, and SVM for myoelectric hand control.
- To identify optimal feature selection strategies for two-electrode systems.
Main Methods:
- Acquired sEMG signals from nine participants performing six hand actions using a two-electrode Delsys Trigno system.
- Applied signal processing techniques and machine learning algorithms (LDA, KNN, SVM).
- Analyzed classification accuracy, action-specific accuracy, and F1-score, exploring feature-accuracy relationships.
Main Results:
- Achieved overall classification accuracy of 93 ± 2% and action-specific accuracy of 97 ± 2%.
- SVM algorithm outperformed LDA and KNN, yielding an F1-score of 87 ± 7%.
- Classification accuracy showed a logarithmic relationship with the number of features, plateauing at five features.
Conclusions:
- Two-electrode sEMG systems can achieve high classification accuracies for myoelectric control.
- SVM offers superior performance for this application compared to LDA and KNN.
- Findings support improved signal processing and machine learning strategies for widely used two-electrode myoelectric hand systems.
