Performance Comparison of Classification Methods for Surface EMG-Based Human-Machine Interface

Abstract

Reliable control of assistive devices through surface electromyography (sEMG) based human-machine interfaces (HMIs) requires accurate classification of multi-channel sEMG. The design of effective pattern classification methods is one of the main challenges for sEMG-based HMIs. In this paper, the authors compared comprehensively the performance of different linear and nonlinear classifiers for the pattern classification of sEMG with respect to three pairs of upper-limb motions (i.e., hand close vs. hand open, wrist flexion vs. wrist extension, and forearm pronation vs. forearm supination). A feature selection approach based on information gain was also performed to reduce the muscle channels. Overall, the results showed that the linear classifiers produce slightly better classification performance, with or without the muscle channel selection.