Classification of Electromyography Signals Using Support Vector Machine

Abstract

In this paper, a classifier has been designed using Support Vector Machine (SVM) to classify Electromyography (EMG) signals. Given the EMG signals, the SVM-based classifier aims to classify ten individual and combined fingers motion command into one of the predefined set of movements. Prior to classification, EMG data is segmented with a sliding window technique and time domain features such as Mean Absolute Value (MAV), Root Mean Square (RMS), Integrated Average Value (IAV), Waveform Length (WL) and autoregressive model (4th order) are extracted for each window and combined to a feature set. Extracted features are used as inputs to the classification system. A linear SVM (one-against-one method) is used for the multiclass classification of EMG signals. Several window sizes that affect the classification performance have been reported. The best feature set that ensures maximum discrimination between the finger movements has also been reported. Validation shows that support vector machine can classify EMG signals correctly with a higher classification rate suitable for designing prosthetic and assistive devices.