Robust muscle force prediction using NMFSEMD denoising and FOS identification

Area of Science:

• Biomedical Engineering
• Signal Processing
• Biomechanics

Background:

• Surface electromyography (sEMG) signals are crucial for understanding muscle activity.
• Accurate sEMG-based muscle force prediction is challenging due to signal noise and complexity.
• Existing methods often struggle with aliasing noise and require robust prediction models.

Purpose of the Study:

• To develop a robust technique for sEMG and muscle force prediction.
• To introduce an effective method for denoising sEMG signals, specifically addressing aliasing noise.
• To enhance the accuracy of muscle force prediction models using processed sEMG data.

Main Methods:

• A novel non-negative matrix factorization screening empirical mode decomposition (NMFSEMD) was developed for sEMG signal denoising.
• The NMFSEMD method screens noisy intrinsic mode functions (IMFs) from empirical mode decomposition (EMD) results.
• A system identification-based muscle force prediction model using a fast orthogonal search (FOS) and candidate functions was constructed.

Main Results:

• The NMFSEMD method improved signal-noise ratio (SNR) by approximately 15.0 dB.
• The energy percentage (EP) of denoised signals exceeded 90.0%.
• The proposed muscle force prediction model reduced mean square error (MSE) by at least 1.2% compared to traditional and LSSVM-based methods.

Conclusions:

• The proposed NMFSEMD technique effectively denoises sEMG signals, significantly improving SNR and EP.
• The developed system identification-based muscle force prediction model offers enhanced accuracy.
• This integrated approach provides a robust solution for sEMG and muscle force prediction in biomechanical applications.