Single Muscle Surface EMGs Locomotion Identification Module for Prosthesis Control

Abstract

Surface EMG (sEMG) signals along with pattern recognition algorithms demonstrate a significant potential to identify and predict human motor activity. We propose a single-channel sEMG signalbased continuous locomotion identification method using a simple classifier. The performance of the proposed method was evaluated for three daily-life locomotion modes on a dataset of 15 subjects. A ranking-based feature selection method was applied to optimize the feature vector. The performance of the proposed method was compared comprehensively with intuitive feature vectors and principle component analysis (PCA). The mean top performances were shown by the Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Neural Network (NN) classifiers as 98.65 ± 0.23, 98.42 ± 0.68, and 99.41 ± 0.51%, respectively (P > 0.05). Further, the subjectwise performance of individually trained classifiers (5 subjects) was accessed through the performance indices, namely classification accuracy, precision, sensitivity, specificity, and F-score. The obtained results indicated no significant degradation and difference in the performance among subjects (P > 0.05). The encouraging results of the proposed method justify its possible use for efficient prosthesis control.