Using a convolutional neural network features to EMG signals classification with continuous wavelet transformation and LS-SVM .

Abstract

The various hand EMG signal grasps are classified in this study. Because EMG signals offer
critical information about muscle activity, they are commonly used as input to electro muscular
control systems. Each muscle performs a specific function in each movement. Electromyography is
a medical, healthcare, and human-machine interaction diagnostic technique for acquiring an EMG
signal (MMI). e most important component of the locomotion system is the muscular system.
Accordingly, sensors were developed to detect the movement system and diagnose the
electromyogram. Nowadays, While maintaining a modest size, it has improved and become more
accurate. In this paper, The EMG signals are converted into images using CWT, then the EMG
images features are extracted based on convolutional neural network (CNN) , and finally, the EMG
features are categorized by an LS-SVM classifier in Matlab. The main objective of this study is to
classify grasps into six basic hand movements: (1) cylindrical, (2) palm, (3) lat (4) sphere(5) Tip,
and (6) Hook. Finally, electrophysiological patterns of each movement were extracted by extracting
features from the images using CNN where EMG images are divided into (70 percent ) training and
(30 percent ) validation, and then these features are fed into classification using the least square
support vector machine. It produced an accuracy of 94.81%.
Keywords: :