Non-negative matrix factorization for separation of extensor indicis and extensor digiti minimi surface electromyograms
Résumé
In a previous study the instantaneity hypothesis for source separation of extensor indicis (EI) and extensor digiti minimi (EDM) muscles has been tested. It turned out electrode location can impact source instantaneity on the sensors and affect blind source separation (BSS) techniques. When dealing with surface electromyograms (SEMG) enveloppes instead of raw SEMG the possible non instantaneity of the sources on the sensors can be modelled as an additive residuals term also accounting for additive noise. Results of source separation using the independent component analysis (ICA) algorithm JADE were compared to the results using non negative matrix factorization (NMF) on SEMG enveloppes. Using NMF on SEMG enveloppes turned out to be much more efficient and more robust to the instantaneity hypothesis.