EMG Data Augmentation for Grasp Classification Using Generative Adversarial Networks

Abstract

Electromyography (EMG) has been used as an interface for the control of robotic hands for decades but with the improvement of embedded electronics and decoding algorithms, many applications are now envisaged by companies. Deep learning has shown the possibility to increase decoding performance but it requires large amounts of data to show its full capabilities. However, recording such amounts of EMG signals face several issues since recording hours of data from patients is very time-consuming and can result in muscle fatigue. We explore a deep learning data augmentation strategy using generative adversarial networks (GANs) to create high-quality synthetic data to increase the performance of grasp classification.