A Novel Non-Euclidean Adaptive Descriptor for Limb Motion Intent Decoding in EMG-Pattern Recognition System.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electromyogram (EMG) based pattern recognition (PR) is vital for motor intent decoding.
• Existing feature extraction methods have limitations in decoding performance, especially with training-test drift and noise.
• Robustness against factors like white Gaussian noise (WGN) is critical for real-world applications.

Purpose of the Study:

• To propose an unsupervised feature extraction scheme for improved EMG-based motor intent decoding.
• To address the challenge of drift between training and testing datasets.
• To enhance the robustness of myoelectric systems against noise.

Main Methods:

• An unsupervised feature extraction scheme was developed, utilizing a feature adaptation approach based on Riemann operations.
• Drift between training and test sets was minimized by projecting test data to align with training set distributions.
• The method was evaluated for motor intent decoding of 13 hand and finger movements and tested for robustness against WGN.

Main Results:

• The proposed feature extraction technique achieved high performance in motor intent decoding, with an average accuracy of 93.17±2.07%.
• The method demonstrated stable and superior decoding performance against WGN compared to state-of-the-art techniques.
• The scheme effectively reduced drift between training and test sets.

Conclusions:

• The proposed unsupervised feature extraction scheme significantly enhances motor intent decoding performance and robustness.
• This technique can improve the overall reliability of myoelectric systems in commercial and clinical settings.
• The findings have implications for better control of prosthetic devices and assistive technologies, improving quality of life.