Combining Temporal and Spectral Features for Improving Hand Gesture Classification from sEMG Signals

Robotic prosthetic hands are being developed to restore some individuals’ capacity to manipulate objects in their daily lives. Currently, more focus is being placed on applying machine learning algorithms to enhance the dexterity of myoelectric prostheses; due to their high precision and sophisticated control capabilities. These algorithms aim to identify and classify hand gestures from a surface electromyography (sEMG), which will be used later as a command of an object. In this work, we suggest a method for categorizing different hand gestures. One of the advantages of this approach is that it can perform well based on a few features from both the time and frequency domains. We use these features with two well-known machine learning algorithms in the literature: Support Vector Machine (SVM) and K-Nearest Neighbors (KNN). Subsequently, we assess how well these algorithms perform using a multi-channel sEMG dataset comprising hand gestures simulated by 40 participants, including movements such as the neutral state, wrist extension and flexion, grip, ulnar and radial deviation, abduction and adduction of all fingers, supination, and pronation. Four sEMG signal channels are used to record these gestures. SVM attained an accuracy of 95.94%, whereas KNN achieved 95.5%, according to the average accuracy rates derived from 40 subjects and 10 hand movements.