A real-time EMG pattern recognition system based on linear-nonlinear feature projection for a multifunction...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Rehabilitation Robotics

Background:

• Myoelectric control systems translate muscle electrical activity (electromyogram - EMG) into device commands.
• Accurate and timely pattern recognition is crucial for intuitive prosthetic limb control.
• Existing methods face challenges in feature extraction, dimensionality reduction, and real-time processing.

Purpose of the Study:

• To develop a novel real-time electromyogram (EMG) pattern recognition system for controlling a multifunction myoelectric hand.
• To improve the class separability and recognition accuracy of EMG features for enhanced prosthetic control.
• To validate the system's applicability in real-time virtual hand control with minimal operational delay.

Main Methods:

• Utilized a four-channel EMG signal acquisition system.
• Employed Wavelet Packet Transform (WPT) for feature vector extraction from EMG signals.
• Implemented a linear-nonlinear feature projection combining Principal Component Analysis (PCA) for dimensionality reduction and Self-Organizing Feature Map (SOFM) for nonlinear mapping.
• Used a Multilayer Perceptron (MLP) as the final classifier.

Main Results:

• The proposed linear-nonlinear feature projection significantly improved class separability and recognition accuracy compared to relying solely on the classifier's ability.
• PCA effectively reduced feature dimensionality, simplifying the classifier and decreasing processing time.
• SOFM created a new feature space with enhanced class separability.
• The complete real-time control system, including virtual hand operation, was achieved within 125 ms.

Conclusions:

• The developed EMG pattern recognition method, incorporating WPT, PCA, and SOFM, is highly effective for real-time multifunction myoelectric hand control.
• The linear-nonlinear feature projection strategy is key to achieving high recognition accuracy and overcoming limitations of traditional methods.
• The system demonstrates practical applicability for prosthetic hand control without operational time delays, paving the way for more intuitive human-machine interfaces.