Deep Learning-Based Classification Using Cumulants and Bispectrum of EMG Signals

EUGENIO OROSCO; JEREMIAS GAIA; JAVIER GIMENEZ; CARLOS M SORIA

IEEE LATIN AMERICA TRANSACTIONS

IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC

Lugar: New York; Año: 2019

1548-0992

Surface electromyographic signals (EMG) histori-cally have been used to classify tasks in basis of a featureextraction scheme and low complexity classifiers. Deep networks,as Multilayer Perceptron and Convolutional Neural Network(MLP and CNN, respectively), avoid the traditional, complexand heuristic (handcrafted) process of feature extraction. Today,it is possible to face the computational cost that these automatictechniques require due to the technology advancement. Thisallowed deep learning techniques to be quickly generalized tocountless applications. This paper proposes to use the thirdorder cumulants and their 2D Fourier transform (Bispectrum)to directly feed CNN and MLP deep learning networks. Theclassifier is not user-dependent (same classifier for all users)and obtains better results than the classical scheme accordingto several metrics.