Subclass Discriminant Analysis based Myoelectric Hand Motion Recognition
Title: Subclass Discriminant Analysis based Myoelectric Hand Motion Recognition
Authors: Dalin Zhou (School of Computing, University of Portsmouth); Yinfeng Fang (Hangzhou Dianzi University); Zhaojie Ju (School of Computing, University of Portsmouth); Honghai Liu (School of Computing, University of Portsmouth);
Year: 2019
Citation: Zhou, D., Fang, Y., Ju, Z., Liu, H., (2019). Subclass Discriminant Analysis based Myoelectric Hand Motion Recognition. UK-RAS19 Conference: “Embedded Intelligence: Enabling & Supporting RAS Technologies” Proceedings, 121-124. doi: 10.31256/UKRAS19.33
Abstract:
Control of prosthetic hands and other upper-limb assistive device for rehabilitation relies on the premise that users’ hand motion intention is accurately recognised. Among all the feasible modalities, myoelectric hand motion recognition has been most adopted yet suffers from its intrinsic day-to-day changes. Despite the promising accuracy achieved by pattern recognition approaches in intra-day tests, the inter-day performance deteriorates in long-term use. From the users’ perspective, it is desired that the hand motion recognition accuracy improves while the burden of user training is confined within 1 or 2 days. In this paper, subclass discriminant analysis is applied instead of conventional linear discriminant analysis for myoelectric hand motion recognition for long-term use. The evaluation results on 10 days’ myoelectric data captured from 6 subjects show that the subclass division contributes to improved inter-day recognition accuracy with limited training data.