[The application of BP neural network improved with LM algorithm in surface EMG signal classification]

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Surface electromyography (sEMG) signals are crucial for prosthetic limb control.
• Accurate classification of sEMG signals representing different movements is a significant challenge.
• Existing methods may lack the speed and precision required for real-time prosthetic applications.

Purpose of the Study:

• To propose an improved Backpropagation (BP) neural network model for enhanced sEMG signal classification.
• To leverage the Levenberg-Marquardt (LM) algorithm to optimize the BP neural network's performance.
• To achieve accurate identification of four distinct forearm movements for prosthetic control.

Main Methods:

• Utilized wavelet transform for data reduction and preprocessing of sEMG signals.
• Implemented a BP neural network enhanced with the Levenberg-Marquardt (LM) algorithm.
• Trained and tested the classifier on four classes of forearm movements: hand extension, hand grasp, forearm pronation, and forearm supination.

Main Results:

• The improved BP neural network with the LM algorithm demonstrated high accuracy in classifying the four forearm movements.
• Significant increases in both the speed and accuracy of sEMG signal classification were observed.
• The method proved effective in practical applications for prosthetic control.

Conclusions:

• The LM-improved BP neural network offers a superior approach for sEMG signal classification compared to standard methods.
• This advancement holds significant potential for improving the dexterity and responsiveness of prosthetic devices.
• The proposed method enhances the feasibility of real-time control for upper-limb prosthetics.