Comparison of Hand Gesture Classification from Surface Electromyography Signal between Artificial Neural Network and Principal Component Analysis

Abstract

The goal of this research is to detect Surface Electromyography (SEMG) signal from
a person’s arm using Myo Armband and classify his / her performed finger ges-tures based on
the corresponding signal. Artificial Neural Network (based on the machine learning approach)
and Principal Component Analysis (based on the feature extraction approach) with and without
Fast Fourier Transform (FFT) were selected as the methods utilized in this research. Analysis
results show that ANN has achieved 62.14% gesture classifying accuracy, while PCA without
FFT has achieved 30.43% and PCA without FFT has achieved 48.15% accuracy. The three
classifiers are tested using SEMG data from a set of six recorded custom gestures. The
comparison results show that the ANN classifier shows higher classifying accuracy and more
robust rather than the PCA classifier’s classi-fying accuracy. Therefore, ANN classifier is more
suited to be implemented in classifying SEMG signals as hand gestures.