A Scalogram Image-Based Method for Continuous Estimation of Hand Kinematics from Surface Electromyography

Area of Science:

• Biomedical Engineering
• Neuroscience
• Robotics

Background:

• Accurate estimation of hand kinematics from surface electromyography (sEMG) is vital for controlling multi-degree-of-freedom (multi-DoF) robotic neural interfaces.
• Current methods face limitations in accuracy, stability, and generalization due to incomplete neural coding models and insufficient information mining.
• Developing advanced techniques is crucial for extracting motor information from neural signals for improved human-computer interaction.

Purpose of the Study:

• To present a novel scalogram image-based method for continuous estimation of multi-DoF finger joint angles using sEMG signals.
• To overcome the limitations of existing methods in accuracy, stability, and generalization ability for hand kinematics estimation.
• To enable more effective control of robotic neural interfaces through enhanced motor information extraction.

Main Methods:

• Applied Continuous Wavelet Transform (CWT) to sEMG signals to generate scalogram images.
• Utilized digital image processing (DIP) for dimensionality reduction of the scalogram images.
• Employed a Convolutional Neural Network (CNN) for end-to-end autonomous learning on the processed images, validated on the Ninapro DB2 dataset.

Main Results:

• The proposed scalogram image-based CNN method achieved a high average correlation coefficient (CC) of 0.9793 ± 0.0089.
• Demonstrated a low average normalized root mean square error (nRMSE) of 0.0490 ± 0.0073, indicating superior estimation accuracy.
• Outperformed control methods, including a CNN with sEMG images (sEMGimage-CNN) and a CWT-based Support Vector Regression (CWT-SVR) method.

Conclusions:

• The developed method offers an effective approach for the continuous estimation of hand kinematics from sEMG signals.
• This research provides a new perspective for myoelectric control, paving the way for more natural human-computer interaction (HCI).
• The findings suggest significant potential for practical applications in advanced prosthetic and robotic control systems.