Multi-scale EMG classification with spatial-temporal attention for prosthetic hands.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Rehabilitation Technology

Background:

• Electromyography (EMG) signals are crucial for controlling prosthetic limbs.
• Accurate hand gesture classification is essential for intuitive prosthetic control.
• Existing methods may not fully capture the complex spatio-temporal dynamics of EMG signals.

Purpose of the Study:

• To develop and evaluate a novel classification framework for hand gestures using EMG signals.
• To leverage multi-scale features and spatial-temporal attention for improved classification accuracy.
• To enhance the performance and reliability of myoelectric prosthetic hands.

Main Methods:

• Utilized a Convolutional Neural Network (CNN) architecture.
• Incorporated a multi-scale coarse-grained layer for enhanced feature extraction from 1D-CNN.
• Implemented a spatial-temporal attention mechanism for feature refinement.
• Classified hand gestures based on processed EMG signal features.

Main Results:

• Achieved high classification accuracies across multiple datasets: 93.4% (Ninapro DB1), 92.8% (DB2), 91.3% (DB5), and 94.1% (DB7).
• Demonstrated the effectiveness of the multi-scale feature extraction and attention mechanism.
• The proposed framework shows significant potential for real-world prosthetic applications.

Conclusions:

• The developed EMG-based hand gesture classification framework offers high accuracy and robustness.
• The integration of multi-scale features and spatial-temporal attention significantly improves classification performance.
• This advancement can lead to more intuitive and reliable control of prosthetic hands, boosting user confidence.