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
MS-ResSNN: A Multi-Scale Residual Spiking Neural Network for Electromyography Pattern Recognition
Summary
A novel multi-scale residual spiking neural network (MS-ResSNN) improves electromyography (EMG) pattern recognition for gesture classification. This advanced spiking neural network (SNN) enhances information propagation in multilayer networks, achieving high accuracy.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Neuroscience
Background:
- Electromyography (EMG) pattern recognition is crucial for human-computer interaction applications like prosthetic control.
- Spiking neural networks (SNNs) offer low power consumption for EMG pattern recognition but face information propagation limitations in multilayer architectures.
- Existing SNN methods often yield suboptimal gesture recognition performance due to restricted information flow.
Purpose of the Study:
- To introduce a novel multi-scale residual spiking neural network (MS-ResSNN) to overcome information propagation challenges in multilayer SNNs for EMG pattern recognition.
- To enhance gesture recognition accuracy by improving information transmission across SNN layers.
- To evaluate the efficacy of the proposed MS-ResSNN for high-density EMG (HD-sEMG) based gesture classification.
Main Methods:
- Developed a multi-scale residual spiking neural network (MS-ResSNN) incorporating residual structures and multi-scale sets of Leaky Integrate-and-Fire (LIF) neurons.
- Optimized spiking encoding and feature extraction using a T-loop network structure for HD-sEMG data.
- Integrated encoded information with extracted EMG features to facilitate enhanced information transmission.
Main Results:
- The MS-ResSNN achieved a high average accuracy of 94.06% on a dataset of 34 gestures from 20 subjects.
- Demonstrated significantly superior performance compared to baseline methods, including spiking convolutional neural networks (SCNN) and convolutional neural networks (CNN).
- The proposed method effectively addressed the information propagation limitations inherent in multilayer SNNs for EMG pattern recognition.
Conclusions:
- The MS-ResSNN presents a significant advancement in EMG pattern recognition, particularly for gesture classification.
- The novel architecture effectively enhances information transmission in multilayer SNNs, leading to improved performance.
- This study highlights the potential of advanced SNNs for robust and accurate gesture recognition using HD-sEMG signals.
