Investigation of Pattern Recognition System Based On Electromyography Signals for Optimal Electrodes’ Number and Positions
This paper proposes a pattern recognition system for classification of six hand movements and rest by using only three Surface ElectroMyoGraphy (sEMG) sensors with the use of Arduino microcontroller as data collector. The performance of the Time Domain with Auto Regression (TDAR) and the recently proposed Time-Dependent Power Spectral Descriptors (TD-PSD) were compared as feature extraction and the k-Nearest Neighbor (KNN) and Linear Discriminant Analysis (LDA) algorithms as classifiers. In addition, the effect of electrodes’ location on forearm and effect of channels number on performance of the pattern recognition system were investigated. Results showed that the performance of the TD-PSD and LDA is higher than that of TDAR and KNN where good classification accuracy was achieved by using only three channels (sEMG) which represented the best three electrode locations for recognizing the six hand movements and rest. Classification accuracy of 97 % was achieved by using only three sEMG channels using low cost components like Arduino and Myoware sEMG sensors which make the proposed system is low cost.