DiffSpkSync: A Muscle Synergy-Guided Spiking Diffusion Model for EMG Signal Generation to Improve Gesture Recognition
IEEE Journal of Biomedical and Health Informatics
|April 30, 2026
Summary
This study introduces DiffSpkSync, a new method to generate more training data for high-density surface electromyography (HD-sEMG) hand gesture recognition (HGR). This approach enhances HGR model accuracy, crucial for medical and rehabilitation applications.
Area of Science:
- Biomedical Engineering
- Machine Learning
- Human-Computer Interaction
Background:
- High-density surface electromyography (HD-sEMG) is vital for intuitive human-machine interaction.
- Scarcity of training data limits the performance of HD-sEMG based hand gesture recognition (HGR) models, particularly in medical and rehabilitation contexts.
Purpose of the Study:
- To address data scarcity in HD-sEMG based HGR.
- To propose DiffSpkSync, a novel generative framework to improve HGR model training data.
Main Methods:
- DiffSpkSync integrates muscle synergy-guided diffusion modeling for signal reconstruction.
- It employs spiking neuron-based sparsification for reduced energy consumption.
- A time-series mixup strategy is used to preserve local dynamics during data augmentation.
Main Results:
- Training gesture classifiers with DiffSpkSync augmented data consistently improved classification accuracy in both intrasession and intersession scenarios.
- DiffSpkSync outperformed representative generative baselines like VAE, DCGAN, DANN-CRC, and PatchEMG.
- Real-time validation showed an average end-to-end latency of 130.22 ms and 95.87% accuracy.
Conclusions:
- DiffSpkSync effectively enhances HGR model performance by addressing data scarcity.
- The proposed framework demonstrates superior results compared to existing generative methods.
- The method's efficiency and accuracy support its application in real-world HGR systems.
