Actor-Critic Reinforcement Learning Based Algorithm for Contaminant Minimization in sEMG Movement Recognition.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface electromyography (sEMG) signals are crucial for movement recognition but are susceptible to various noise types.
• Common contaminants include electrocardiography (ECG) interference, motion artifacts (MOA), powerline interference (PLI), and additive white Gaussian noise (WGN).
• Effective noise reduction is vital for reliable sEMG-based motion classification systems.

Purpose of the Study:

• To develop and evaluate a Reinforcement Learning (RL) based approach for detecting and mitigating noise in sEMG signals.
• To enhance the accuracy of movement recognition systems by applying the RL method during the pre-processing stage.
• To assess the performance of the RL approach against different types and levels of signal contamination.

Main Methods:

• An RL-based algorithm was developed for the pre-processing of sEMG data.
• The method was tested on the Ninapro database 2, artificially contaminated with ECG, MOA, PLI, and WGN.
• Movement classification was performed using a Support Vector Machine (SVM) classifier.

Main Results:

• The RL pre-processing algorithm demonstrated significant improvements in movement recognition accuracy.
• Accuracy gains ranged from 8.9% for one contaminated channel to 16.7% for three, 15.9% for six, 16.5% for nine, and 11.9% for 12 contaminated channels.
• The results indicate the effectiveness of the RL approach in restoring contaminated sEMG signals.

Conclusions:

• The proposed RL-based pre-processing method effectively detects and minimizes noise in sEMG signals.
• This approach leads to substantial improvements in the accuracy of sEMG-based movement recognition.
• The RL technique offers a promising solution for enhancing the robustness of prosthetic and assistive devices relying on sEMG data.