Autores: Orosco, E.; di Sciascio, F.;
Resumen: High-order statistics (HOS) are well suited for describing non-Gaussian random processes. These techniques are increasingly being employed in myoelectric research, on both time and frequency domain techniques. This work presents HOS-based techniques using only HOS time domain features to classify myoelectric signals. The auto-, cross- and full- (joint) third-order cumulants are evaluated as EMG-signal feature vectors to be compared between them. Four surface EMG signals were processed for classify motions from the upper limbs. Synergy among channels is characterized by the features in both auto and cross modes, and their incidences for classifying five or six movements are analyzed. In contrast to the third-order auto-cumulants, it had been verified that the third-order cross-cumulants have the same classification rate by working with five or six movements. A myoelectric control scheme and its experimental application were executed with normal and disabled subjects, reaching a classification rates of 90%, in average. Accuracy in online experiments was similar to the off-line classification rate.