An inventive quadratic time-frequency scheme based on Wigner-Ville distribution for classification of sEMG signals

Electromyogram signal is a biopotential signal that may be measured on the surface of contracting muscles representing neuromuscular activities. This signal may be utilized in various applications such as clinical diagnosis of diseased neuromuscular systems and as a measurement tool for evaluation of rehabilitation activities. Another recent application is the usage of EMG signal in design and implementation of neural controlled prosthesis hands. For this purpose appropriate features of EMG signal are required such that intended hand movements may be recognized correctly. In this work we considered a new method based on quadratic time-frequency representation namely Wigner-Ville distribution (WVD) to extract required information. In the proposed approach, initially WVD coefficients for each class were calculated. Next average coefficients for all the signals in each class were obtained. Then cross-WVD was found by using acquired average WVD coefficients with signals in each class and finally the number of zero crossing (ZC) of cross-WVD coefficients were utilized as suitable features. Our proposed approach provided satisfactory results with a recognition average accuracy rate of 91.3% for six classes of movements. On the other hand, for unprocessed (raw) WVD coefficients the average accuracy of the six hand movements was registered at %33.7.