Improved myoelectric pattern recognition of finger movement using rejection-based extreme learning machine

Abstract

Myoelectric control system (MCS) had been applied to hand exoskeleton to improve the human-machine interaction. The current MCS enables the exoskeleton to move all fingers concurrently for opening and closing hand and does not consider robustness issue caused by the condition not considered in the training stage. This study addressed a new MCS employing novel myoelectric pattern recognition (M-PR) to handle more movements. Furthermore, a rejection-based radial-basis function extreme learning machine (RBF-ELM) was proposed to tackle the movements that are not included in the training stage. The results of the offline experiments showed the RBF-ELM with rejection mechanism (RBF-ELM-R) outperformed RBF-ELM without rejection mechanism and other well-known classifiers. In the online experiments, using 10-trained classes, the M-PR achieved an accuracy of 89.73% and 89.22% using RBF-ELM-R and RBF-ELM, respectively. In the experiment with 5-trained classes and 5-untrained classes, the M-PR accuracy was 80.22% and 59.64% using RBF-ELM-R and RBF-ELM, respectively