Correlation between the stability of feature distribution and classification performance in sEMG signals

Research output: Chapter in Book/Report/Conference proceedingConference paperpeer-review

3 Citations (Scopus)
36 Downloads (Pure)

Abstract

The long-term robustness of pattern recognition-based myoelectric systems draws more attention from researchers. Though, there is a lack of analysis investigating how features change over time. This study used two metrics: Coefficient of variation of the first four moments (CoV) and Two-Sample Kolmogorov-Smirnov Test statistics (K-S); to quantify the stability of feature distributions and correlate their changes over time to classification performance. We acquired two surface electromyography (sEMG) channels from sixteen subjects (ten able-bodied and six trans-radial amputees) performing three hand motions. Results showed that the selected metrics correlate to some degree to classification accuracy. Feature distributions are affected less by the time when data are combined. These results imply that stable temporal change may be an acceptable way to choose robust features in long term investigations.
Original languageEnglish
Title of host publicationBioSMART 2021 - Proceedings
Subtitle of host publication4th International Conference on Bio-Engineering for Smart Technologies
PublisherIEEE
ISBN (Electronic)978-1-6654-0810-3
Publication statusPublished - 21 Jan 2022
Event2021 4th International Conference on Bio-Engineering for Smart Technologies (BioSMART) - Paris / Créteil, Paris, France
Duration: 8 Dec 202110 Dec 2021

Publication series

NameBioSMART 2021 - Proceedings: 4th International Conference on Bio-Engineering for Smart Technologies

Conference

Conference2021 4th International Conference on Bio-Engineering for Smart Technologies (BioSMART)
Country/TerritoryFrance
CityParis
Period8/12/202110/12/2021

Keywords

  • Long-term
  • Pattern recognition
  • Surface Electromyography
  • Feature
  • Myoelectric control

Fingerprint

Dive into the research topics of 'Correlation between the stability of feature distribution and classification performance in sEMG signals'. Together they form a unique fingerprint.

Cite this