A Graph Neural Network Model for Real-Time Gesture Recognition Based on sEMG Signals.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human-Computer Interaction

Background:

• Accurate hand gesture recognition is crucial for controlling advanced prosthetics and augmented reality systems.
• Surface electromyography (sEMG) signals from forearm muscles are widely used for this purpose.
• Existing methods face challenges in real-time accuracy and efficiency.

Purpose of the Study:

• To develop a novel approach for sEMG signal representation using graph networks.
• To create a machine learning algorithm for real-time hand gesture recognition based on these graph networks.
• To evaluate the performance and efficiency of the proposed method.

Main Methods:

• Utilized graph networks to represent muscle activation patterns from forearm sEMG signals.
• Developed a graph neural network (GNN) algorithm for real-time hand gesture recognition.
• Evaluated the algorithm using sEMG data from 8 healthy subjects with a myoband device.

Main Results:

• Achieved an average classification accuracy of 99% for hand gesture recognition.
• Demonstrated superior performance compared to state-of-the-art techniques.
• Attained an average processing time of 48ms for graph construction and prediction on an M1 Pro CPU.

Conclusions:

• The proposed graph network-based approach offers highly accurate and efficient real-time hand gesture recognition.
• This method is well-suited for seamless control of advanced prosthetic devices and augmented reality applications.
• The novel sEMG representation significantly enhances the performance of gesture recognition systems.