EMG Signals based Human Action Recognition via Deep Belief Networks

Abstract

Electromyography (EMG) signals can be used for action classification. Nonetheless, due to their nonlinear and time-varying properties, it is difficult to classify the EMG signals and it is critical to use appropriate algorithms for EMG feature extraction and classification. In previous studies various ML methods have been applied. In this paper, we extract four time-domain features of the EMG signals and use a generative graphical model, Deep Belief Network (DBN), to classify the EMG signals. A DBN is a fast, greedy deep learning algorithm that can find a set of optimal weights rapidly, even in deep networks with many hidden layers and a large number of parameters. To evaluate this model, we acquired EMG signals, extracted their features, and then utilized the DBN model as human action classifiers. The real data analysis results are presented to show the effectiveness of the proposed deep learning technique for 4-class recognition of human actions based on the measured EMG signals. The proposed DBN model has potential to be applied in design of EMG-based user interfaces.