On the Use of Temporal and Spectral Central Moments of Forearm Surface EMG for Finger Gesture Classification

Analyzing the surface electromyogram (sEMG) signal is becoming increasingly popular in fields other than medical diagnostics, such as assistive technology and human machine interfaces. This work focusses on analysing data from three sEMG sensors placed on the forearm in an armband configuration for the purpose of identification of finger gestures in a sign-language recognition system. The higher order central moments defining the shape of the power spectral density (PSD) are found to be particularly useful for the considered application. A comparative study of temporal and spectral central moments derived from the probability density function (PDF) and PSD of sEMG signals, respectively, is carried out to study their utility in the aforementioned application. Practical experiments reveal that spectral moments along with the most prominently used set of features out-perform the temporal moments in the considered classification. An average classi?cation accuracy of 82.1% is achieved with temporal moments, which is improved to 90.1% with spectral moments.