Related Experiment Videos
Feature-based classification of myoelectric signals using artificial neural networks
P J Gallant1, E L Morin, L E Peppard
1Department of Electrical & Computer Engineering, Queen's University, Kingston, Ontario, Canada.
Medical & Biological Engineering & Computing
|April 13, 1999
Summary
This study presents a pattern classification system to differentiate myoelectric signals based on muscle contraction tasks. The system utilizes artificial neural networks (ANNs) to analyze signal structures for improved classification accuracy.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Machine Learning
Background:
- Myoelectric signals contain task-specific information.
- Advanced pattern recognition can classify these signals.
- Existing methods may not fully capture signal complexity.
Purpose of the Study:
- To describe a novel pattern classification system for myoelectric signals.
- To leverage signal structure for differentiating contraction tasks.
- To evaluate the system's performance and complexity.
Main Methods:
- Spectrographic preprocessing to generate power spectral densities.
- Feature extraction using a self-organizing artificial neural network (ANN) with exploratory projection pursuit.
- Classification of extracted features by a supervised-learning ANN.
Main Results:
- The system effectively separates myoelectric signal records based on contraction tasks.
- The amplitude of myoelectric signals within 200 ms of contraction onset shows task-specific non-random structure.
- The described ANN-based approach demonstrates potential for advanced signal analysis.
Conclusions:
- The developed pattern classification system is effective for myoelectric signal analysis.
- Artificial neural networks, particularly with exploratory projection pursuit, offer powerful tools for feature extraction.
- The system's performance and complexity are suitable for practical applications in myoelectric control.