Resilient EMG Classification to Enable Reliable Upper-Limb Movement Intent Detection.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Surface electromyography (sEMG) signals are crucial for controlling assistive devices but are challenging due to their stochastic nature.
• Current methods for sEMG pattern recognition often suffer from classification errors and delays caused by data processing and sample discarding.

Purpose of the Study:

• To develop and evaluate a resilient classification pipeline using Extreme Learning Machines (ELM) for robust real-time control of upper-limb assistive devices.
• To improve the accuracy and consistency of sEMG-based movement classification compared to existing methods.

Main Methods:

• A novel classification pipeline based on Extreme Learning Machines (ELM) was implemented.
• The pipeline was used to classify 17 distinct upper-limb movements from sEMG signals across three different databases.
• Performance was benchmarked against a baseline ELM and a sample discarding (DISC) method.

Main Results:

• The proposed ELM pipeline demonstrated more stable and consistent classifications compared to baseline ELM and DISC methods.
• An average accuracy increase of approximately 10% was observed across all databases.
• Achieved average weighted accuracy rates of over 53.4% for amputees and 89.0% for non-amputee volunteers.

Conclusions:

• The resilient ELM classification pipeline offers a significant improvement in sEMG-based control for assistive devices.
• The method achieves comparable or superior results to existing approaches without the need for sample discarding, reducing processing delays.
• This approach holds promise for more reliable and intuitive real-time control of prosthetic and assistive technologies.