Lower Limb EMG Signal Analysis Using Scattering Transform and Support Vector Machine for Various Walking Conditions

Abstract

Electromyography (EMG)-based clinical signal processing and feature extraction are crucial for diagnosing various diseases. In this work, using lower limb EMG data, scattering transform (ST)-based features and time-domain-based features were compared using a support vector machine (SVM) classifier. Several subjects were instructed to walk on a platform designed to transform its surface between even and uneven terrain. The EMG signals from the gastrocnemius (GM) and medial hamstring muscles (HM) of twenty-one healthy male subjects were acquired under different walking conditions (i.e., even forward, even backward, uneven forward, and uneven backward walking directions). The features were extracted using a time-domain-based method and ST method which were then fed into an SVM classifier. Test results showed that time-domain features can provide accuracy ranges between 97.31 and 99.57% and ST features can provide accuracy ranges between 99.42 and 100%. The high number of extracted features used showed that ST features require a longer processing time compared to time-domain features. In addition, the analysis of generated scatterplots showed how time-domain-based features contribute to the classification of walking conditions. The findings of this study have potential applications in both clinical and bioengineering fields.

Keywords

• Electromyography
• feature extraction
• scattering transform
• support vector machine
• walking gait analysis