Hand Movement Recognition Using Cross Spectrum Image Analysis of EMG Signals-A Deep Learning Approach

Abstract

Accurate recognition of hand movements using electromyography (EMG) signals is important to develop robust human-machine interface system for upper limb prosthetic control. For accurate detection and classification of different movements, proper feature selection from EMG signals is necessary, failure of which may often lead to incorrect results. To this end, this article presents an image processing aided deep learning approach for detection and classification of hand movement EMG signals. In this study, EMG signals for both left hand and right hand movements were procured from an existing database. A reference signal for each category was chosen as reference and cross spectrum of the rest of the EMG signals was done with the reference signal. The resultant cross-wavelet spectrum images of both classes of EMG signals were fed to a pre-trained convolution neural network (CNN) for the purpose of detection and classification of hand movements. It has been observed that the proposed method returned an average classification accuracy of 97.6% in segregating different categories of EMG signals. Besides, the performance of the proposed method analyzed using different CNN architectures was also found to be robust. The proposed method can be implemented for real-time detection of hand movements.