Myoelectric Signal Classification of Targeted Muscles Using Dictionary Learning

Area of Science:

• Biomedical Engineering
• Neuroscience
• Rehabilitation Engineering

Background:

• Surface electromyography (sEMG) signals are crucial for myoelectric prostheses control.
• Efficient processing of sEMG signals is vital, yet optimal algorithms and electrode placements remain under investigation.
• Minimizing electrode count for sEMG signal processing is an underexplored area.

Purpose of the Study:

• To identify the most effective method for myoelectric signal classification using a minimal number of electrodes.
• To compare the classification accuracy of discriminative feature-oriented dictionary learning (DFDL) against conventional classifiers.
• To evaluate the impact of targeted versus untargeted muscle signal acquisition on classification performance.

Main Methods:

• Acquired sEMG data from 23 subjects performing 14 distinct hand movements.
• Collected sEMG data from both targeted and untargeted muscles.
• Compared classification accuracy using DFDL and other standard classifiers, analyzing performance with varying electrode channel counts.

Main Results:

• DFDL achieved the highest classification accuracy among all tested classifiers.
• DFDL's superior performance was more pronounced with a reduced number of electrode channels.
• Targeted muscle signal acquisition significantly outperformed untargeted acquisition, especially when using DFDL.

Conclusions:

• The combination of targeted muscle acquisition and the DFDL algorithm provides superior myoelectric signal classification.
• This approach enables effective classification with a minimal number of channels, advancing prosthetic control.
• The findings suggest a promising direction for developing more efficient and user-friendly myoelectric prosthetic devices.