An Enhanced Random Convolutional Kernel Transform for Diverse and Robust Feature Extraction from High-Density Surface...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Human-Machine Interaction

Background:

• High-density surface electromyogram (HD-sEMG) is crucial for hand gesture recognition.
• Current methods lack feature diversity or require extensive user-specific training due to neuromuscular variations.

Purpose of the Study:

• To introduce EMG-ROCKET, a novel feature extraction method for HD-sEMG.
• To enhance robustness and reduce data dependency in hand gesture recognition models.

Main Methods:

• EMG-ROCKET, an adaptation of ROCKET, utilizes random channel fusion and enhanced aggregation.
• The method extracts diverse and robust HD-sEMG features without prior knowledge or extensive training.
• Evaluated using a Ridge classifier on two HD-sEMG datasets for cross-day hand gesture recognition.

Main Results:

• EMG-ROCKET features achieved 84.3% and 77.8% accuracy in cross-day evaluations, outperforming baseline methods.
• Demonstrated robustness against day-to-day signal variability.
• Feature contribution analysis revealed insights into spatial muscle activation patterns.

Conclusions:

• EMG-ROCKET offers a training-free solution for robust HD-sEMG feature extraction.
• The method enhances hand gesture recognition accuracy and provides insights into neuromuscular mechanisms.
• Facilitates practical applications in human-machine interaction.