Subject Independent Classification of Hand Gesture from sEMG using an Approximate Entropy Based Approach

• Published in: 2020 11th International Conference on Electrical and Computer Engineering (ICECE) pp. 238 - 241
• Main Authors: Paul, Joydip, Alam, Mohammad Tahmidul, Paul, Sudip
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.12.2020
• Subjects: Across subject; Approximate Entropy (ApEn); Automation; Discriminant analysis; Entropy; Performance evaluation; Prosthetics; Surface electromyogram (sEMG); Time-domain analysis; Usability; Within subject
• DOI: 10.1109/ICECE51571.2020.9393127

Surface electromyogram (sEMG) signal is used as a convenient tool in prosthetics because of its accessibility and unobtrusiveness. However, because of the low signal quality, sEMG based prosthesis automation suffers from low accuracy. Besides, the force and neuromuscular stimulation related to a movement varies from person to person and results in poor generalizability. Most of the existing hand gesture classification methods propose subject-specific models that don't have satisfactory accuracy across subject cases and hence lose their real-life usability. In this paper, we proposed a method based on approximate entropy (ApEn) that achieves high classification accuracy for both within and across subject cases. The method consists of a feature set combining ApEn with five time-domain (TD) and frequency domain (FD) features and it was tested on a publicly available sEMG dataset having six different hand movements of five subjects. A six-class linear discriminant analysis confirms that the method achieves above 96% accuracy and 70-80% accuracy within and across subject cases respectively. Because of high accuracy in subject independent tests, the method promises to perform well in automatic prosthetic devices in real-life scenarios.