Removing Placement Artifact of sEMG with Generative Adversarial Network based Domain Adaptation for Robotic Applications

Abstract

Surface electromyography(sEMG) is widely used sensor to get human intention of motion. Most of modern prosthesis includes sEMG sensors to move it. However, sEMG is not aggressively applied in commercial prosthesis yet. It is because there are many problems unsolved. One of the problem is in-consistent sensor placement. We call it placement artifact. Although, in many studies, it was shown that sEMG pattern recognition can extract human intentions very well, the fact that pattern recognition would fail if it is used with different sensor placement, is not considered in the most studies. But it is critical problem and should be solved to apply sEMG to prosthesis control with stable performance. To manage this problem, we suggests a method for removing placement artifact. This method utilizes Transfer Learning and Generative Adversarial Network(GAN) framework. Transfer Learning is the research area seeking a method for transferring the knowledge in one domain or task to another domain or task. In this work, we suggests sensor placement measure and used it as an domain indicator. i.e. different sensor placement measure means different domain. We trained GAN network that can transfer sEMG data in one domain to another domain. As an result, we could remove placement artifact through trained network.