Spectral model based intent detection for multichannel SEMG signals
- Resource Type
- Conference
- Authors
- Patil, Reena; Kang, Ke; Ozturk, Yusuf
- Source
- 2017 IEEE EMBS International Conference on Biomedical & Health Informatics (BHI) Biomedical & Health Informatics (BHI), 2017 IEEE EMBS International Conference on. :469-472 2017
- Subject
- Bioengineering
Engineering Profession
Computational modeling
Feature extraction
Wrist
Electromyography
Data models
Brain modeling
Training
- Language
This study proposes a multichannel spectral model for detection and classification of upper limb motion. Multichannel surface electromyography (SEMG) signals are segmented into active and idle regions before computing model coefficients. Autoregressive (AR) model coefficients computed over the active regions are used for characterization and classification of SEMG signals. Itakura-Satio spectral distance (ISD) measure is used during the classification for comparing AR coefficients of the selected segments and the reference AR models. The method proposed here successfully classified 7 different upper limb motions with an accuracy of 92.71%. Model based approach achieved a better classification accuracy than most traditional feature based approaches at a lower complexity and increased number of motion classes.