Hand Gesture Classification Using Surface Electromyography Signals Based on Fusion of Time Domain Features

Surface electromyogram (sEMG) decoders that en-able hand gesture recognition have garnered more interest due to the growing demands on human-machine interfaces in intelligent natural rehabilitation. Although previous studies have proposed diverse models for gesture classification using sEMG signals, their recognition performance degrades due to the following reasons: (i) the extracted features are not distinguishable to recognize closely related gestures, (ii) with an increase in feature dimension, the recognition model gets confused to classify the correct class, (iii) the gesture classification performance degrades for larger number of gesture class. In this work, several distinguished hand-crafted features including waveform length (WL), mean absolute value (MAV), standard deviation (SD), and root mean square (RMS) features, are derived from each channel of the sEMG signals to define a gesture class. The feature combinations are evaluated using variety of machine learning models. The proposed features and SVM classifier outperform prior reported findings on Ninapro-DB5 standard dataset, with an accuracy of 84%.