Continuous estimation of multi-DOF movement from sEMG based on non-negative matrix factorization and L2 regulation

Area of Science:

• Biomedical Engineering
• Robotics
• Signal Processing

Background:

• Accurate continuous estimation of multi-degree-of-freedom (DOF) movement is essential for advanced myoelectric prosthetic control.
• Decoupling multi-DOF movements presents a significant challenge in achieving precise continuous estimation.

Purpose of the Study:

• To propose and validate a novel model for suppressing non-active DOF and achieving continuous multi-DOF movement estimation.
• To enhance the control capabilities of myoelectric prosthetics through improved movement decoding.

Main Methods:

• A model combining non-negative matrix factorization (NMF) with Hadamard product and L2 regularization was developed.
• L2 regularization was applied to the non-active DOF activation coefficient within the NMF objective function.
• Movement angles were estimated via a linear combination of calculated activation coefficients.

Main Results:

• The proposed NMF-based model demonstrated superior suppression of non-active DOF in single-DOF movements compared to existing muscle synergy methods.
• Experiments validated the model's effectiveness in continuous estimation for both single-DOF and multi-DOF wrist and hand movements.
• The study confirmed the robustness of the suppression effect and the consistency of synergy matrices across different movement speeds.

Conclusions:

• The novel NMF model effectively suppresses non-active DOF, enabling accurate continuous estimation of multi-DOF movements.
• This approach offers a significant advancement for the simultaneous control of sophisticated myoelectric prosthetics.
• The method shows promise for improving prosthetic functionality and user control.