Feature Extraction of Surface Electromyography Based on Improved Small-World Leaky Echo State Network

摘要

Surface electromyography (sEMG) is an electrophysiological reflectionof skeletal muscle contractile activity that can directly reflect neuromuscularactivity. It has been a matter of research to investigate feature extractionmethods of sEMG signals. In this letter, we propose a featureextraction method of sEMG signals based on the improved small-worldleaky echo state network (ISWLESN). The reservoir of leaky echo statenetwork (LESN) is connected by a random network. First, we improvedthe reservoir of the echo state network (ESN) by these networks andused edge-added probability to improve these networks. That idea enhancesthe adaptability of the reservoir, the generalization ability, andthe stability of ESN. Then we obtained the output weight of the networkthrough training and used it as features. We recorded the sEMG signals during different activities: falling, walking, sitting, squatting, goingupstairs, and going downstairs. Afterward, we extracted correspondingfeatures by ISWLESN and used principal component analysis fordimension reduction. At the end, scatter plot, the class separability index,and the Davies-Bouldin indexwere used to assess the performance of features.The results showed that the ISWLESN clustering performance wasbetter than those of LESN and ESN. By support vector machine, it wasalso revealed that the performance of ISWLESN for classifying the activitieswas better than those of ESN and LESN.