Continuous Motion Estimation of Knee Joint Based on a Parameter Self-Updating Mechanism Model

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Signal Processing

Background:

• Estimating continuous human joint motion from surface electromyography (sEMG) is vital for intelligent rehabilitation.
• Traditional models struggle with individual differences and signal non-stationarity, limiting generalizability.
• Accurate sEMG-based motion estimation is crucial for advancing human-robot interaction.

Purpose of the Study:

• To develop a robust continuous motion estimation model for the human knee joint.
• To address the challenges of non-stationarity and individual variability in sEMG signals.
• To enhance the generalizability of sEMG-based joint motion estimation methods.

Main Methods:

• A novel model fusing particle swarm optimization (PSO) and deep belief network (DBN) was proposed.
• The method features a parameter self-updating mechanism for adaptive DBN optimization.
• High-dimensional signal feature reconstruction was employed for optimal estimation.

Main Results:

• The proposed PSO-DBN model achieved average root mean square errors (RMSEs) of 9.42° and 7.36° for knee joint motion.
• Performance surpassed that of common neural network approaches in experimental evaluations.
• The method demonstrated effective adaptive optimization and feature reconstruction.

Conclusions:

• The developed PSO-DBN model significantly improves the accuracy of continuous knee joint motion estimation from sEMG.
• This approach offers enhanced generalizability across different subjects.
• The findings provide a foundation for advanced human-robot interaction in exoskeleton robotics.