Simple space-domain features for low-resolution sEMG pattern recognition.

Area of Science:

• Biomedical Engineering
• Rehabilitation Engineering
• Human-Computer Interaction

Background:

• Low-cost myoelectric control systems (e.g., Myo armband) offer new possibilities but face limitations due to low sampling frequencies of surface electromyography (sEMG) signals.
• The validity of existing sEMG feature extraction methods on low-resolution data is questionable, and algorithms must be computationally efficient for embedded systems.

Purpose of the Study:

• To investigate effective features for low-resolution EMG pattern recognition.
• To develop novel, computationally efficient space-domain (SD) features for sEMG analysis.
• To evaluate the performance of these features on a real-world myoelectric control device.

Main Methods:

• Development of simple space-domain (SSD) features that exploit spatial relationships in sEMG signals from sensor arrays.
• Evaluation of SSD features using a linear discriminant analysis (LDA) classifier.
• Testing on a 9-gesture dataset using the Myo armband.

Main Results:

• The proposed SSD features demonstrated effectiveness in low-resolution sEMG pattern recognition.
• Classification accuracy was improved by 5% compared to traditional Hudgins' time-domain features.
• SSD features offer a computationally efficient alternative for embedded myoelectric control.

Conclusions:

• Simple space-domain (SSD) features are a viable and effective method for enhancing myoelectric control accuracy with low-resolution sEMG data.
• These findings support the use of SSD features in real-time, embedded myoelectric control applications.
• The developed features provide a practical solution for overcoming limitations of low-frequency sEMG sampling.