Deep Learning-Based Identification Algorithm for Transitions Between Walking Environments Using Electromyography...

Area of Science:

• Biomedical Engineering
• Robotics
• Human-Computer Interaction

Background:

• Existing terrain identification for walking assistive devices often relies on sensor fusion.
• Studies specifically classifying gait transitions using only electromyography (EMG) signals are limited.

Purpose of the Study:

• To propose an identification algorithm for transitions between various walking environments using only EMG signals.
• To leverage deep learning for classifying gait environment changes based on lower extremity muscle activity.

Main Methods:

• Measured EMG signals from 27 subjects across multiple lower extremity muscles during walking on diverse terrains (flat, stairs, slopes) and transitions.
• Utilized an artificial neural network (ANN) model, inputting the entire EMG profile during the stance phase.
• Evaluated classification accuracy using all muscle activations and a subset of key muscle groups.

Main Results:

• The ANN model achieved a high classification accuracy of 95.4% for identifying transitions between walking environments using all measured muscle activations.
• A reduced set of muscle activations (knee extensor, ankle extensor, metatarsophalangeal flexor) yielded a classification accuracy of 90.9%.

Conclusions:

• Transitions between different gait environments can be accurately identified using only EMG signals during the stance phase.
• The developed ANN model demonstrates the potential for precise gait transition classification, paving the way for advanced walking assistive devices.