Detection of Hand Movement Using Time, Frequency and Time-Frequency Features of the Electromyogram Signal in Order to Create a Human-Machine Interface

This study presents an electromyography (EMG)-based hand gesture recognition system that leverages machine learning techniques to accurately classify hand gestures. A total of 33 participants (21 men and 12 women, average age 24.8 years) performed six hand movements under controlled conditions. EMG signals were recorded at a sampling rate of 1024 Hz and filtered using a band-pass Butterworth filter (20–450 Hz) and a notch filter to remove power-line interference. Signal normalization ensured consistency across trials, and extensive feature extraction was conducted, capturing time-domain, frequency-domain, time-frequency, and nonlinear characteristics. To optimize classification, three feature selection methods were employed: Principal Component Analysis (PCA), Recursive Feature Elimination (RFE), and Mutual Information (MI). Results indicate that MI-based feature selection, coupled with a Radial Basis Function (RBF) Support Vector Machine (SVM) classifier, provided the highest accuracy ($\mathbf{9 1 . 7 \%}$) across key metrics, outperforming linear SVM. This study highlights the importance of signal preprocessing, feature extraction, and appropriate classifier selection for enhancing the robustness of EMG-based gesture recognition. The findings have potential applications in prosthetics, assistive technology, and human-machine interfaces (HMIs), offering a reliable approach for interpreting complex EMG signals.