Compound motion decoding based on sEMG consisting of gestures, wrist angles, and strength

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Signal Processing

Background:

• The need for sophisticated control of electromyographic (EMG) prosthetics is increasing.
• Current EMG-based control systems often lack the dexterity for complex upper limb movements.
• Decoding compound motions from surface EMG (sEMG) is crucial for versatile prosthetic operation.

Purpose of the Study:

• To evaluate and compare deep learning and machine learning approaches for decoding upper limb compound motions.
• To identify optimal feature sets and classifiers for high-accuracy sEMG-based motion decoding.
• To enhance the flexibility and diversity of control for EMG-driven hand prosthetics.

Main Methods:

• Selected 60 compound motions, varying gestures, wrist angles, and strength levels.
• Compared deep learning models (three structures, two label encodings) with machine learning classifiers (24 classifiers, seven features, classifier chains).
• Utilized a feature combination including mean absolute value, root mean square, variance, autoregressive coefficient, wavelength, zero crossings, and slope signal change.

Main Results:

• The Support Vector Machine (SVM) with a quadric kernel and the selected feature combination achieved superior performance.
• Achieved high average test accuracy of 98.42 ± 1.71% for compound motion decoding using 150 ms sEMG.
• Individual accuracies for gestures (99.35%), wrist angles (99.34%), and strength levels (99.04%) were exceptionally high, with 58 out of 60 motions exceeding 95% accuracy.

Conclusions:

• Feature combination with SVM provides a highly accurate, stable, and computationally efficient solution for multi-target sEMG classification.
• This approach significantly improves the decoding of complex upper limb movements for prosthetic applications.
• The developed method offers a promising pathway for more intuitive and functional EMG-controlled prosthetic devices.