A sEMG Classification Framework with Less Training Data.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Surface electromyograph (sEMG) signals are used with machine learning for muscle state classification.
• Artificial Neural Networks (ANN) are common but require extensive training data.

Purpose of the Study:

• To develop a novel framework for a binary sEMG classifier.
• To distinguish repetitive dumbbell motion using sEMG data.
• To reduce the amount of training data required for sEMG classification.

Main Methods:

• A new framework for designing a binary sEMG classifier was introduced.
• The framework leverages prior knowledge of sEMG signals.
• The classifier's performance was validated through experiments.

Main Results:

• The proposed sEMG classifier achieved a 95.7% success rate in distinguishing user states.
• This accuracy is comparable to ANN classifiers (99.6%) but with significantly less training data.
• Under identical training conditions, the proposed framework outperformed ANN, which dropped to 65.6% accuracy.

Conclusions:

• The novel framework effectively designs a high-accuracy sEMG classifier.
• This approach reduces the need for extensive training data collection.
• The method shows promise for efficient muscle state classification in real-world applications.