Transient state analysis of the multichannel EMG signal using Hjorth's parameters for identication of hand movements
Résumé
Abstract--Most myoelectric controlled systems are based on the common assumption that there is no information in the instantaneous value of the myoelectric signal and therefore, analysis is made on the steady state of the muscle contraction. However, two main drawbacks of this control scheme are that the user needs to be trained in order to produce the sustained contractions, and that the control signal can only be generated when the steady state is reached. Using the transient state would allow classifying the movement during the dynamic part of the contraction decreasing the delay and allowing a more natural control. This paper proposes a novel method for transient EMG classification based on the use of Hjorth's parameters. Surface multichannel EMG signals were recorded from 10 normally limbed subjects for both the transient and steady EMG states. Six different hand motions were accurately classified using a window length of 128 ms and an artificial neural network model. An average classification accuracy of 97.45% (SD 1.10%) was obtained for the transient EMG state and of 97.93% (SD 1.11%) for the steady state, suggesting the existence of deterministic information in the transient state of the muscle contraction and the fact that Hjorth's parameters seem to adapt well enough to the nature of myoelectric signals as to extract highly representative information from them.