Related Experiment Video
Updated: Aug 5, 2026
08:15
Capturing Dynamic Finger Gesturing with High-resolution Surface Electromyography and Computer Vision
Published on: March 28, 2025
Spatiotemporal Dual-Channel Interpretable Hybrid Neural Network for HD-sEMG-Based Gesture Recognition
Zhefei Cai1,2,3, Su Liu2,4,5,6, Xinyue Li2
1College of Information Engineering, China Jiliang University, Hangzhou 310018, China.
Sensors (Basel, Switzerland)
|July 28, 2026
Summary
We developed an interpretable deep learning model for high-density surface electromyography (HD-sEMG) to improve prosthetic control. Our STDC-Net achieves high accuracy in gesture recognition, offering better insights into feature importance and channel interactions.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Neuroscience
Background:
- Accurate gesture recognition is vital for controlling upper limb prostheses.
- High-density surface electromyography (HD-sEMG) offers rich data for improved myoelectric control.
- Deep learning enhances HD-sEMG gesture recognition but often lacks interpretability.
Purpose of the Study:
- To develop an interpretable hybrid neural network (STDC-Net) for HD-sEMG-based gesture recognition.
- To improve the interpretability and optimization of deep learning models in prosthetic control.
- To validate STDC-Net's performance and interpretability using the Capgmyo DB-a dataset.
Main Methods:
- Developed a spatiotemporal dual-channel interpretable hybrid neural network (STDC-Net).
- Utilized Feature Channels and Spatial Channels for signal processing and interpretability.
- Employed SHapley Additive exPlanations (SHAP) for feature importance ranking and graph attention layers for channel relationship analysis.
Main Results:
- STDC-Net achieved 99.8% accuracy for intra-subject tasks and 97.33 ± 2.53% for inter-subject tasks, outperforming SOTA methods.
- The model exceeded real-time implementation requirements for prosthetic control.
- SHAP value maps and channel connection maps provided detailed insights into feature contributions and network interactions, enhancing interpretability.
Conclusions:
- STDC-Net demonstrates superior performance and enhanced interpretability for HD-sEMG-based gesture recognition.
- The model's interpretability aids feature filtering and understanding of parameter interactions.
- STDC-Net shows significant promise for advancing real-time prosthetic control systems.