The Use of Time and Frequency Features in Finger Movements Based on Electromyogram Recording

Abstract

Human movement is a direct result of brain commands to muscles. The muscle activity can be analyzed with an electromyograph device and produce a recording in the form of an electromyogram or EMG signal. There are two recording methods, namely intramuscular EMG and surface EMG or sEMG which are easier and more convenient to implement but have lower resolution. Analysis of the EMG record data will be better if the right method is chosen, one of which is the selection of feature extraction methods. This study performs basic finger motion classification with a limited number of electrodes, namely 4 electrodes. The purpose of this research is to compare the variation of features in the time or frequency domain that will be used for classification using artificial neural networks (ANN) and long short-term memory (LSTM). Data acquisition was carried out with the Ganglion Board device on 6 subjects aged 20–22 years who conducted experiments with two kinds of movements, namely opening and holding all fingers alternately. The results of this study indicate that the use of time domain features for classification with ANN produces better accuracy than LSTM. This can happen because the duration of the movement in this study is quite short, which is only for two seconds. The results of the use of the frequency domain feature show that the use of LSTM produces better accuracy, especially on the mean power and median frequency characteristics.