Optimizing Feature Reduction and Selection Techniques for Surface Electromyography

Abstract

A biological signal called an EMG monitors the electric impulses produced in muscles at various points during their contraction, which are indicative of neuromuscular processes. Surface electromyography’s (sEMG) extraction of features is a substantial procedure to obtain the valuable information that is obscure in the data and to exclude redundant components and intrusions. Feature vector selection is crucial for effective EMG signal classification. However, several redundant characteristics have been found in studies of the categorization of EMG signals when utilizing a particular feature set. This research has discussed the properties and characteristics of different frequency and time domain features. Many features of the time domain are unnecessary and redundant and can be categorized based on the information and mathematical qualities, according to the scatter plots of different features, arithmetical analysis, or different classifiers. Contrarily, the statistical characteristics of the EMG spectral density of power are used to determine all frequency domain properties. The EMG recognition system's performance at the class separability level is inappropriate. This research also suggests using inefficient features for classifications in applications involving the classification of EMG signals