Feature Selection and Non-Linear Classifiers: Effects on Simultaneous Motion Recognition in Upper Limb.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Robotics

Background:

• Myoelectric signals (EMG) are vital for volitional control of prosthetic limbs.
• Effective feature selection is crucial for accurate motion classification from EMG data.
• Simultaneous motion classification in prosthetics requires robust signal processing techniques.

Purpose of the Study:

• To evaluate feature selection methods for classifying simultaneous upper limb motions from EMG signals.
• To identify optimal features and algorithms for prosthetic control applications.
• To provide recommendations for implementing motion classification from EMG signals.

Main Methods:

• Utilized Chow-Liu trees and forward feature selection for identifying relevant EMG signal features.
• Evaluated six different classification algorithms, including linear and non-linear kernel approaches.
• Analyzed classification accuracy based on selected features and algorithms.

Main Results:

• Feature selection significantly improved classification accuracy to over 95%.
• Chow-Liu trees offered comparable accuracy to forward selection with fewer iterations.
• Waveform length and entropy emerged as the most relevant feature types.
• Non-linear classification approaches proved superior for simultaneous motion classification.

Conclusions:

• Optimized feature and algorithm selection enhances surface EMG-based simultaneous motion classification for upper limb prosthetics.
• The study provides a viable approach for recognizing combined upper limb actions.
• Recommendations are established for improved EMG signal processing in prosthetic control.