A Novel Unsupervised Adaptive Learning Method for Long-Term Electromyography (EMG) Pattern Recognition.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Long-term use of pattern recognition-based myoelectric control systems leads to performance degradation due to various interfering factors.
• Adaptive learning methods are crucial for maintaining system performance over time, especially in unsupervised scenarios.

Purpose of the Study:

• To propose a low-computational-cost adaptive learning method to mitigate performance degradation in myoelectric control.
• To introduce the particle adaptive classifier (PAC) for unsupervised adaptive learning in myoelectric control applications.

Main Methods:

• Developed a particle adaptive classifier (PAC) integrating a particle adaptive learning strategy and a universal incremental least square support vector classifier (LS-SVC).
• Compared PAC performance against incremental support vector classifier (ISVC) and non-adapting support vector classifier (NSVC).
• Evaluated classification performance and retraining time cost using simulated and realistic long-term electromyography (EMG) data in both unsupervised and supervised adaptive learning settings.

Main Results:

• PAC significantly reduced performance degradation in unsupervised adaptive learning on realistic long-term EMG data compared to NSVC (9.03% ± 2.23%, p < 0.05) and ISVC (13.38% ± 2.62%, p = 0.001).
• PAC demonstrated a substantial reduction in retraining time cost compared to ISVC (2 ms vs. 50 ms per updating cycle).
• Classification accuracy was validated across simulated and realistic datasets.

Conclusions:

• The proposed particle adaptive classifier (PAC) effectively mitigates performance degradation in long-term myoelectric control, particularly in unsupervised adaptive learning.
• PAC offers a computationally efficient solution for maintaining the accuracy and reliability of myoelectric control systems.
• The findings suggest PAC is a promising approach for enhancing the robustness of prosthetic limb control and other myoelectric applications.