Classification of Walking Environments Using Deep Learning Approach Based on Surface EMG Sensors Only.
Pankwon Kim1, Jinkyu Lee1, Choongsoo S Shin1
1Department of Mechanical Engineering, Sogang University, 35 Baekbeom-ro, Mapo-gu, Seoul 04107, Korea.
Sensors (Basel, Switzerland)
|July 2, 2021
Summary
This study demonstrates that surface electromyography (sEMG) signals alone can accurately classify walking environments. An artificial neural network achieved 96.3% accuracy using sEMG data from lower extremity muscles.
Area of Science:
- Biomechanics
- Robotics
- Biomedical Engineering
Background:
- Terrain classification is crucial for advanced control of walking assistive devices.
- Existing methods often combine surface electromyography (sEMG) with other sensors, but classifying environments using only sEMG is unexplored.
- Accurate terrain recognition enhances the adaptability and safety of lower extremity assistive technologies.
Purpose of the Study:
- To classify different walking environments solely based on sEMG signals from lower extremity muscles.
- To evaluate the efficacy of an artificial neural network (ANN) in terrain classification using sEMG.
- To determine which muscle groups provide the most significant contribution to accurate terrain classification.
Main Methods:
- Collected sEMG data from 27 participants walking on five different terrains: flat-ground, upstairs, downstairs, uphill, and downhill.
- Utilized an artificial neural network (ANN) to classify walking environments based on the complete sEMG profile during the stance phase.
- Analyzed classification accuracy using all muscles collectively and individual muscle groups (knee, ankle, metatarsophalangeal flexors/extensors).
Main Results:
- The ANN achieved a high classification accuracy of 96.3% when utilizing sEMG activation from all measured lower extremity muscles.
- Individual muscle group analysis revealed that the triceps surae muscle activation yielded the highest classification accuracy of 88.9%.
- The findings confirm the potential of sEMG as a standalone sensor for robust terrain classification.
Conclusions:
- Walking environments can be accurately classified using only sEMG signals and an ANN.
- sEMG-based terrain classification offers a promising, sensor-independent approach for intelligent walking assistive devices.
- This method enhances the potential for more intuitive and responsive control of mobility aids across diverse terrains.
