Feasibility of NeuCube spiking neural network architecture for EMG pattern recognition

Abstract

Multichannel electromyography (EMG) signals have been used as human-machine interface (HMI) for the control of pattern-recognition based prosthetic system in recent years. This paper is a feasibility analysis of using recently proposed NeuCube spiking neural network (SNN) architecture for a 6-class recognition problem of hand motions. NeuCube is an integrated environment, which uses SNN reservoir and dynamic evolving SNN classifier. NeuCbube has the advantage of processing complex spatio-temporal data. The preliminary experiments show that Neucube is more efficient for EMG classification than commonly used machine learning techniques since it achieves better accuracy as well as consistent classification outcomes. The performance of NeuCube combined with TD features reaches up to 95.33% accuracy after a careful selection of the features. This paper demonstrates that NeuCube has the potential to be employed in practical applications of myoelectric control.