Attenuating the impact of limb position on surface EMG pattern recognition using a mixed-LDA classifier

Surface electromyography (sEMG) signals have been applied in the control of prosthetic hand over the past decades. With many control schemes and algorithms being proposed, pattern recognition (PR) based control strategies have become a mainstream approach for the manipulation of dexterous prostheses...

Full description

Saved in:
Bibliographic Details
Published in2017 IEEE International Conference on Robotics and Biomimetics (ROBIO) pp. 1497 - 1502
Main Authors Yu, Yang, Sheng, Xinjun, Guo, Weichao, Zhu, Xiangyang
Format Conference Proceeding
LanguageEnglish
Japanese
Published IEEE 01.12.2017
Subjects
Acceleration
Electromyography
Feature extraction
limb position variation
mixed-LDA classifier
Pattern recognition
Sensors
surface electromyography
Testing
Training data
Online AccessGet full text
DOI10.1109/ROBIO.2017.8324629

Cover

More Information
Summary:Surface electromyography (sEMG) signals have been applied in the control of prosthetic hand over the past decades. With many control schemes and algorithms being proposed, pattern recognition (PR) based control strategies have become a mainstream approach for the manipulation of dexterous prostheses. However, there is still a huge gap between practical usage and controlled laboratory conditions due to electrode shift, consistence during contraction, impedance variation and so on. Among these challenges, the variation of limb position between training session and test session is an important issue which will have bad impact on the motion classification performance. In this paper, we investigated the impact of limb position variation on the PR based motion classification with a linear discrimination analysis (LDA) classifier. Experiments were carried out in five different limb positions and seven classes of hand motion in each position. In order to reduce the impact, a mixed-LDA classifier which fused the LDA parameters of other positions was implemented. The results revealed that mixed-LDA classifier performed better than single-set classifier with a 93.6% classification accuracy over the five limb positions. The outcome of this study suggests that mixed-LDA method may be a promising way to attenuate the adverse impact of limb position variation.
DOI:10.1109/ROBIO.2017.8324629