Swarm-based extreme learning machine for finger movement recognition

Publication Type:
Conference Proceeding
Middle East Conference on Biomedical Engineering, MECBME, 2014, pp. 273 - 276
Issue Date:
Filename Description Size
ThumbnailMECBME_ANAM.pdf Published version313.58 kB
Adobe PDF
Full metadata record
An accurate finger movement recognition is required in many robotics prosthetics and assistive hand devices. The use of a small number of Electromyography (EMG) channels for classifying the finger movement is a challenging task. This paper proposes a novel recognition system which employs Spectral Regression Discriminant Analysis (SRDA) for dimensionality reduction, kernel-based Extreme Learning Machine (ELM) for classification and the majority vote for classification smoothness. Particle Swarm Optimization (PSO) is used to optimize the kernel-based ELM. Three hybridizations with three kernels, radial basis function (SRBF-ELM), linear (SLIN-ELM), and polynomial (SPOLY-ELM) are introduced. The experimental results show that SRBF-ELM significantly outperforms SLIN-ELM but not too much different compared to SPOLY-LIN. Moreover, PSO is able to optimize the three systems by giving the accuracy more than 90% with the highest accuracy is ∼94%. © 2014 IEEE.
Please use this identifier to cite or link to this item: