EMGFlow: Robust and Efficient Surface Electromyography Synthesis via Flow Matching

2026-04-15

AI Summary

This work addresses the limitations of surface electromyography (sEMG)-based gesture recognition, which suffers from data scarcity and insufficient subject diversity, compounded by suboptimal training stability or inference efficiency in existing generative approaches. To overcome these challenges, the authors propose EMGFlow—the first conditional generative framework to introduce Flow Matching into the sEMG domain. EMGFlow leverages continuous-time generative modeling, high-order numerical solvers, and a tailored temporal sampling strategy to achieve high-fidelity signal generation while significantly accelerating inference. Evaluated under a unified Train-on-Synthetic, Test-on-Real (TSTR) protocol, EMGFlow consistently outperforms conventional data augmentation techniques, GANs, and diffusion models across multiple benchmark datasets, demonstrating a superior trade-off between generation quality and computational efficiency as well as strong practical utility for downstream tasks.

Technology Category

• Natural Language Processing: Generation
• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision

Application Category

• Web Mining and Content Analysis: Web data generation and simulation
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems

Abstract

Deep learning-based surface electromyography (sEMG) gesture recognition is frequently bottlenecked by data scarcity and limited subject diversity. While synthetic data generation via Generative Adversarial Networks (GANs) and diffusion models has emerged as a promising augmentation strategy, these approaches often face challenges regarding training stability or inference efficiency. To bridge this gap, we propose EMGFlow, a conditional sEMG generation framework. To the best of our knowledge, this is the first study to investigate the application of Flow Matching (FM) and continuous-time generative modeling in the sEMG domain. To validate EMGFlow across three benchmark sEMG datasets, we employ a unified evaluation protocol integrating feature-based fidelity, distributional geometry, and downstream utility. Extensive evaluations show that EMGFlow outperforms conventional augmentation and GAN baselines, and provides stronger standalone utility than the diffusion baselines considered here under the train-on-synthetic test-on-real (TSTR) protocol. Furthermore, by optimizing generation dynamics through advanced numerical solvers and targeted time sampling, EMGFlow achieves improved quality-efficiency trade-offs. Taken together, these results suggest that Flow Matching is a promising and efficient paradigm for addressing data bottlenecks in myoelectric control systems. Our code is available at: https://github.com/Open-EXG/EMGFlow.

Problem

• surface electromyography
• data scarcity
• subject diversity
• synthetic data generation
• myoelectric control

Innovation

• Flow Matching
• surface electromyography
• generative modeling
• synthetic data augmentation
• myoelectric control