Improving Fast EMG Classification for Hand Gesture Recognition: A Comprehensive Analysis of Temporal, Spatial, and...

Area of Science:

• Biomedical Engineering
• Neurorehabilitation
• Signal Processing

Background:

• Assistive devices using electromyography (EMG) show promise for restoring mobility in post-stroke individuals.
• Real-time functionality of myoelectric control systems is challenged by biological signal variability and processing delays.

Purpose of the Study:

• To evaluate electromyography (EMG) signal classification performance for hand gestures in healthy and post-stroke individuals.
• To analyze the impact of acquisition time and channel count on myoelectric control models.
• To identify optimal feature extraction and machine learning methods for robust real-time control.

Main Methods:

• Classification of six distinct hand gestures using electromyography (EMG) signals.
• Analysis of feature extraction methods including power spectral density and dimensionality reduction.
• Evaluation of acquisition times (0.5-4 s) and channel numbers (1-4) on model performance.
• Testing model generalization using intra-patient and cross-patient validation on post-stroke data.

Main Results:

• Achieved high classification accuracy (up to 95.31%) with power spectral density and dimensionality reduction.
• Determined that acquisition time (stabilizing at 2 s) impacts accuracy more than the number of channels.
• Demonstrated intra-patient validation accuracy of 90% and cross-patient validation of 35-40% for generalization.

Conclusions:

• Optimized EMG signal processing enhances the accuracy and robustness of myoelectric control systems.
• Findings contribute to the development of effective real-time control for neurorehabilitation devices.
• Further research is needed to improve cross-patient generalization for broader clinical application.