Mitigating the Concurrent Interference of Electrode Shift and Loosening in Myoelectric Pattern Recognition Using...

Area of Science:

• Biomedical Engineering
• Human-Computer Interaction
• Machine Learning

Background:

• Myoelectric pattern recognition (MPR) systems are crucial for intelligent control but are sensitive to electrode displacement.
• Electrode shift and loosening cause significant interference, limiting the practical application of MPR-based gestural interfaces.

Purpose of the Study:

• To develop a novel MPR method robust against concurrent electrode shift and loosening.
• To enhance the practicality and reliability of MPR for intelligent control systems.

Main Methods:

• A Siamese auto-encoder network (SAEN) was developed to learn feature representations resistant to electrode interference.
• The SAEN was trained using simulated shifted-view and masked-view feature maps with three mean square error (MSE) losses.
• The SAEN functioned as a feature extractor, followed by a support vector machine classifier.

Main Results:

• The proposed SAEN-based MPR method achieved the highest classification accuracy under concurrent interference conditions in both offline and online tests.
• Statistical significance (p < 0.05) was observed compared to five other common methods.
• The method demonstrated effectiveness in mitigating interferences caused by electrode shift and loosening.

Conclusions:

• The novel SAEN-based MPR method significantly improves robustness against common electrode interferences.
• This approach offers a valuable solution for enhancing the reliability of myoelectric control systems.
• The findings contribute to the advancement of practical MPR applications in intelligent gestural interfaces.