Recurrent Neural Network for Contaminant Type Detector in Surface Electromyography Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface Electromyography (sEMG) signals are crucial for prosthetics and diagnostics.
• Signal contamination from noise and artifacts can compromise sEMG data integrity.
• Accurate contaminant identification is essential for reliable sEMG analysis.

Purpose of the Study:

• To develop and evaluate a novel contaminant type detector for sEMG signals.
• To leverage Recurrent Neural Networks (RNNs) with Long Short-Term (LSTM) units for direct sEMG signal classification.
• To reduce processing time by eliminating the need for manual feature extraction.

Main Methods:

• Utilized the publicly available NINAPro database for sEMG signals.
• Trained and tested a Long Short-Term Memory (LSTM) based RNN model.
• Introduced contaminants including White Gaussian Noise, Movement Artifact, ECG, and Power Line Interference.
• Trained 12 models, one per channel, under -20dB contamination levels.

Main Results:

• Achieved an overall classification accuracy of 97.72% with one of the trained models.
• Demonstrated that the training channel can influence classification accuracy at specific Signal-to-Noise Ratios (SNRs).
• Confirmed the model's capability to classify different types of sEMG signal contamination directly from raw data.

Conclusions:

• The developed RNN-LSTM model offers an efficient and accurate method for sEMG contaminant detection.
• Direct classification from sEMG signals significantly reduces processing time compared to traditional feature-based methods.
• This approach holds promise for improving the reliability of real-time sEMG-based applications, such as advanced prosthetics.