Related Experiment Video
Updated: Jun 11, 2025
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
407
Hand gesture recognition using sEMG signals with a multi-stream time-varying feature enhancement approach
Jungpil Shin1, Abu Saleh Musa Miah2, Sota Konnai2
1School of Computer Science and Engineering, The University of Aizu, Aizuwakamatsu, Fukushima, 965-0006, Japan. jpshin@u-aizu.ac.jp.
Scientific Reports
|September 27, 2024
Summary
A novel deep learning system enhances surface electromyography (sEMG) for hand gesture recognition. This muscle-computer interface achieves high accuracy, improving prosthetic limb control and human-machine interaction.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Human-Computer Interaction
Background:
- Surface electromyography (sEMG) based hand gesture recognition is crucial for muscle-computer interfaces but faces challenges in performance due to ineffective feature enhancement, leading to unstable predictions.
- Existing sEMG systems struggle with erratic predictions and require improved feature extraction methods for reliable hand gesture recognition.
Purpose of the Study:
- To develop a lightweight and high-performance sEMG-based hand gesture recognition system.
- To address the limitations of existing systems by introducing a novel 4-stream deep learning architecture for enhanced feature extraction and stable predictions.
Main Methods:
- A 4-stream deep learning architecture combining Temporal Convolutional Network (TCN) and Convolutional Neural Network (CNN) modules.
- Utilizing Temporal Convolutional Network (TCN) for time-varying features, a hybrid Long Short-Term Memory (LSTM)-TCN module for long-range temporal relations, and a CNN-TCN merge for spatio-temporal features.
- Incorporating a skip connection mechanism for robust information flow and a channel attention-based feature selection module to reduce computational complexity.
Main Results:
- The proposed model achieved high accuracy, with an average of 94.31% on the Ninapro DB1 dataset and 98.96% on the Ninapro DB9 dataset.
- The 4-stream architecture effectively extracts comprehensive spatio-temporal features, leading to superior performance compared to existing methods.
- The channel attention mechanism successfully identified and selected the most effective features, enhancing model efficiency.
Conclusions:
- The novel lightweight deep learning system significantly improves sEMG-based hand gesture recognition accuracy and stability.
- This advancement holds substantial implications for enhancing the quality of life for individuals using prosthetic limbs.
- The proposed system offers a promising solution for more sophisticated control in robotic human-machine interfaces.
