A Deep Learning Based Personal Authentication Method Using sEMG Signal

• Published in: Chinese Control Conference pp. 6316 - 6320
• Main Authors: Tong, Lina, Zhang, Mingjia, Liu, Daisong, Ma, Hanghang, Zhang, Yuxiang, Peng, Liang
• Format: Conference Proceeding
• Language: English
• Published: Technical Committee on Control Theory, Chinese Association of Automation 25.07.2022
• Subjects: Authentication; Convolution; Deep learning; Electromyography; Law; LSTM Network; multi-layers Convolution Neural Network (CNN); Myo; Neural networks; Personal Authentication; Skin; Surface Electromyography(sEMG)
• ISSN: 1934-1768
• DOI: 10.23919/CCC55666.2022.9901683

Personal authentication means that the relevant system confirms whether the identity of the user is real, legal and unique. A safe and convenient personal authentication method is a core for guarantee the security of our information and property. Surface Electromyography (sEMG), which exists on the surface of the skin, is a good signal to achieve personal authentication, for it is difficult to be extracted and forged. In this paper , a new personal authentication method using forearm sEMG is proposed. First, 8-channel sEMG signals are recorded by Myo armband that is placed on right forearm of 14 subjects who perform the same hand open gesture. Then, two different deep learning models are presented to classify the signals. The multi-layers Convolution Neural Network (ML-CNN) and two stages Long Short Term Memory (2s-LSTM) Network can achieve accuracy of 97.50% and 93.60% respectively, shows it is feasible to achieve the goal of personal authentication. Although the time cost of ML-CNN is higher than the 2s-LSTM, both are lower than human reaction time. So the architecture of ML-CNN is more suitable to achieving the goal of personal authentication.