Feature dimensionality reduction for myoelectric pattern recognition: a comparison study of feature selection and...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Technology

Background:

• Surface electromyography (EMG) signals are crucial for developing myoelectric control systems.
• Dimensionality reduction is essential for efficient and accurate EMG signal classification.
• Current methods often face challenges with redundancy and feature set dependency.

Purpose of the Study:

• To investigate the impact of feature dimensionality reduction strategies on EMG signal classification.
• To develop a practical myoelectric control system using effective feature selection.
• To evaluate feature selection methods independent of classification algorithms and feature types.

Main Methods:

• Two dimensionality reduction strategies, feature selection and feature projection, were applied to EMG feature sets.
• A feature selection based myoelectric pattern recognition system was implemented.
• Markov random field (MRF) and forward orthogonal search algorithms were used for feature evaluation.

Main Results:

• High classification accuracies (>95%) were achieved for seven forearm motions using a small subset of selected EMG features (average 12 features).
• The feature selection approach demonstrated independence from the type of feature set and classification algorithms.
• The method effectively reduced redundant information across channels and within channels.

Conclusions:

• The proposed filter-based feature selection approach enables robust EMG feature dimensionality reduction.
• This method is adaptable and can be integrated with existing classification algorithms, facilitating clinical utility.
• The findings represent a significant step towards practical and reliable myoelectric control systems.