Deep Convolutional Neural Network for Decoding EMG for Human Computer Interaction

Abstract

sEMG is a promising human computer interaction approach, which has been widely used in myriads of areas. To perform sEMG classification, more and more sophisticated machine learning strategies have been developed. However, the deep neural network still has limited applications on sEMG decoding, though it has got a great success in the computer vision area. In this study, we propose a new deep learning framework to classify hand gestures based on sEMG, especially we perform convolutional neural network (CNN) on multiple-session sEMG, which is more challenging because of the time-varying biodynamics of the subjects. So we also investigate the topologies of CNN, expecting to get an optimized architecture to effectively detect the hidden features in the signals. It is shown that the proposed CNN framework in this study has a high classification accuracy for sEMG-based hand gesture recognition, and the difference of topologies has great impact on the performance of CNN. This study lays a promising foundation for multiple-session sEMG signal pattern recognition by CNN.