Classification of Electromyography Signals Using Neural Networks and Features From Various Domains

Abstract

Real-time control of prosthetic hands has attracted huge attention from researchers in recent years. Real-time analysis of Electromyography (EMG) signals has several challenges. The most important one is to achieve an acceptable classification accuracy by observing a limited length of the EMG signal. In this paper, we address these challenges i.e., we enhance the classification accuracy and reduce the required observation signal’s length. These goals are achieved by employing extracted features from time, frequency, and time-frequency domains and introducing a new neural network architecture to combine these features. The experimental results illustrate that combining features from different domains and the proposed architecture improve the accuracy of real-time classification of EMG signals in comparison to existing state-of-the-art methods.