Towards a versatile Surface Electromyography classification system

Abstract

The use of sEMG signals for the movement classification plays an important role in various applications from robotics to effective prosthetic limbs control. The performance of the classification scheme is severely influenced by the efficiency of the used feature set to create discriminant subspaces for each movement. In the recent literature, various feature sets have been proposed, that usually create rather complicated feature spaces. The aim of this research is to propose a versatile scheme based on simple and uniform characteristics capable to significantly improve the performance of the movement classification by using the sEMG signals. The set is comprised of features like energies and a few other features from the well-know and widely used Hudgins set, all estimated on the wavelet domain of the sEMG signal. The application of the proposed scheme on standard database of sEMG signals, the NINAPRO a database that is built for benchmarking and algorithmic evaluation, proved that the classification performance of movements exceeds 96% with a significant improvement when compared with the performance of other schemes proposed. © Springer International Publishing Switzerland 2016.