A novel TCNet for irrelevant gesture rejection based on electromyography signals

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Myoelectric control systems can misinterpret irrelevant gestures, leading to erroneous commands.
• Effective discrimination between intended and irrelevant gestures is crucial for reliable human-machine interaction.

Purpose of the Study:

• To propose a novel temporal neural network, TCNet, for robust myoelectric gesture recognition.
• To address the challenge of irrelevant gesture interference in myoelectric control systems.

Main Methods:

• Utilized Fourier transform to decompose electromyography (EMG) signals into different frequency components, creating 2D representations.
• Employed 2D convolution to learn temporal characteristics within and between signal periods.
• Introduced center loss for improved feature discrimination and a Softmax probability threshold for gesture rejection.

Main Results:

• Achieved a 91.5% recognition rate for 10 target gesture classes.
• Demonstrated a 90.5% rejection rate for 7 irrelevant gesture classes.
• Attained an overall average accuracy of 91.1% across 17 gesture classes from 11 subjects.

Conclusions:

• TCNet effectively mitigates irrelevant gesture interference in myoelectric control.
• The proposed method offers competitive performance compared to existing state-of-the-art approaches.
• This work enhances the reliability and precision of myoelectric control systems.