Recognition of lower limb movements using empirical mode decomposition and k-nearest neighbor entropy estimator with surface electromyogram signals

• Published in: Biomedical signal processing and control Vol. 71; p. 103198
• Main Authors: Wei, Chunfeng, Wang, Hong, Lu, Yanzheng, Hu, Fo, Feng, Naishi, Zhou, Bin, Jiang, Daqi, Wang, Zhen
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2022
• Subjects: Empirical mode decomposition; k-Nearest Neighbor entropy estimator; Lower limb movement recognition; Machine learning; Surface electromyography; Empirical mode decomposition; Surface electromyography; k-Nearest Neighbor entropy estimator; Machine learning; Lower limb movement recognition
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2021.103198

•A novel lower limb movements recognition method is presented by using sEMG signals.•The EMD method is employed to decompose the original sEMG signal.•The KNN-En estimator extracted from IMF has advantages in recognizing lower limb movements.The proposed method only uses sEMG signal sensors and has high accuracy. Lower limb movement recognition is critical to the daily care of the elderly, the weak, and the disabled. Surface electromyogram (sEMG) signals reflect the intention of human movements and can be used as the source of lower limb movement recognition. However, sEMG signals exhibit low stability due to electrode displacement, muscle structure differences, and muscle contraction strength. The effective extraction of features from sEMG signals is considered a difficult problem in the studies on sEMG signals-based lower limb movement recognition. In this work, we proposed a novel method of lower limb motion recognition based on empirical mode decomposition (EMD) and k-Nearest Neighbor entropy (KNN-En) estimator. First, the sEMG signals of four lower limb movements from twenty subjects were recorded with seven wearable sEMG signal sensors, and the sEMG signals were denoised through the multi-scale principal component analysis (MSPCA). Then, the sEMG signals were decomposed by EMD into multiple intrinsic mode functions (IMF), and the KNN-En estimator features were extracted from the IMF. Next, the KNN-En estimator features were projected into low-dimensional spaces by three feature projection techniques, namely principal component analysis (PCA), isometric mapping (Isomap), and diffusion mapping (DM). Finally, the four lower limb movements were recognized by three machine learning classifiers, namely support vector machine (SVM), k-nearest neighbor (KNN), and Bagging. The experimental results showed that the combination of the SVM classifier and the DM method exhibited excellent recognition performance and an accuracy of 99.63%, thereby proving the feasibility of the proposed method in lower limb motion recognition.