Related Experiment Video
Updated: Jun 27, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
Wrist-Wearable sEMG Gesture Recognition System Based on ThinNet Lightweight Neural Network.
Zihao Wang1, Long Meng2, Chen Chen1
1Human Phenome Institute, Fudan University, Shanghai 200433, China.
Bioengineering (Basel, Switzerland)
|June 26, 2026
Summary
We developed a wearable system for surface electromyography (sEMG) gesture recognition, achieving high accuracy with a lightweight neural network. This technology enhances human-machine interaction for practical applications.
Area of Science:
- Biomedical Engineering
- Human-Computer Interaction
- Machine Learning
Background:
- Wearable surface electromyography (sEMG) offers intuitive human-machine interaction but faces challenges like hardware limits, complex models, and user variability.
- Existing systems often struggle with practical deployment due to these limitations.
Purpose of the Study:
- To develop an efficient and accurate wearable sEMG-based gesture recognition system.
- To address hardware constraints and inter-subject variability in sEMG systems.
Main Methods:
- Designed a high-performance wrist-worn sEMG acquisition system with a novel electrode array and filtering.
- Developed ThinNet, a lightweight neural network for gesture recognition.
- Validated the system with 100 participants performing six distinct gestures.
Main Results:
- The sEMG system achieved a high signal-to-noise ratio (SNR) of 66.96 dB.
- ThinNet demonstrated 90.47% inter-subject accuracy, peaking at 96.80% with a buffered strategy.
- The model showed excellent data efficiency, maintaining performance with only 40% fine-tuning data.
Conclusions:
- The combined hardware optimization and lightweight neural network effectively advance wearable sEMG gesture recognition.
- The developed framework is scalable and suitable for practical, real-world applications.
- This approach significantly improves the feasibility of intuitive human-machine interaction using sEMG.
