Improved High-Density Myoelectric Pattern Recognition Control Against Electrode Shift Using Data Augmentation and...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Electrode shift is a major challenge in myoelectric control, leading to reduced accuracy.
• Existing methods struggle to compensate for the spatial variations caused by electrode displacement.
• Robust myoelectric control is crucial for advanced prosthetic devices and human-computer interfaces.

Purpose of the Study:

• To develop a robust myoelectric control method that mitigates the impact of electrode shift.
• To enhance the accuracy and reliability of high-density surface electromyogram (HD-sEMG) pattern recognition.
• To create a practical solution for real-world applications where electrode placement may vary.

Main Methods:

• High-density surface electromyogram (HD-sEMG) signals were preprocessed into image representations.
• A data augmentation technique was employed to simulate various electrode shift positions from a single reference position.
• A dilated convolutional neural network (DCNN) was utilized for classifying myoelectric patterns, leveraging its larger receptive field for spatial context.

Main Results:

• The proposed method achieved a mean classification accuracy of 95.34% under diverse electrode shift conditions.
• The DCNN-based approach demonstrated superior performance compared to other common classification methods.
• The technique effectively handled simulated and actual electrode shifts in HD-sEMG data.

Conclusions:

• The integration of data augmentation and DCNN provides a feasible and usable approach for predicting myoelectric patterns despite electrode shifts.
• This method offers a practical solution for achieving robust myoelectric control in the presence of electrode array displacement.
• The findings highlight the potential of image-based deep learning techniques for overcoming challenges in bio-signal processing.