A Hand Gesture Recognition Strategy Based on Virtual-Dimension Increase of EMG

Area of Science:

• Biomedical Engineering
• Robotics
• Signal Processing

Background:

• Electromyography (EMG) signals control myoelectric bionic hands for gesture recognition.
• Increasing EMG channels improves accuracy but faces diminishing returns, leading to a plateau.
• Existing methods struggle to extract sufficient motion intention from limited physical EMG channels.

Purpose of the Study:

• To propose a method for virtually increasing EMG signal channels to improve gesture recognition accuracy.
• To overcome the information saturation and accuracy plateau in high-density EMG acquisition.
• To introduce a new quantitative measure, separability of feature vectors (SFV), for predicting classification effectiveness.

Main Methods:

• A virtual channel expansion strategy is employed to enrich EMG signal information.
• Feature selection concepts are adapted to develop the SFV measure.
• SFV is calculated based on the divergence and correlation of extracted features.
• The effectiveness of virtual dimension increase is assessed by analyzing changes in feature set differentiability.

Main Results:

• The proposed virtual channel method successfully improves recognition accuracy for various gestures.
• The SFV measure effectively predicts classification performance, outperforming statistical success rates.
• SFV is shown to be a faster, more representative, and suitable measure for small sample sets.

Conclusions:

• Virtual expansion of EMG channels offers a viable solution to the accuracy plateau in bionic hand control.
• The SFV measure provides a robust and efficient tool for evaluating classification effectiveness in feature selection.
• This approach enhances the potential for more intuitive and accurate control of myoelectric prosthetics.