Measuring Neuromuscular Electrophysiological Activities to Decode HD-sEMG Biometrics for Cross-Application Discrepant Personal Identification With Unknown Identities
Xinyu Jiang,Xiangyu Liu,Jiahao Fan,Xinming Ye,Chenyun Dai,Edward A. Clancy,Wei Chen
DOI: https://doi.org/10.1109/tim.2022.3180434
IF: 5.6
2022-10-29
IEEE Transactions on Instrumentation and Measurement
Abstract:Measuring the physical, physiological, behavioral, or chemical characteristics of an individual as biometrics for personal identification has attracted increasing attention in smart environment applications. Noncancelability and cross-application invariance are two flaws of traditional DNA, face, and fingerprint-based biometrics because users cannot volitionally change the biometric template. In this work, we acquired high-density surface electromyogram (HD-sEMG) signals encoded by gesture passwords as biometrics. The different sEMG patterns under different motor tasks allow users to enroll multiple accounts using sEMG under different hand gestures as biometrics. By simply changing to a new gesture password, users can cancel the original template once it is compromised. Even if impostors enter the correct gesture password, the individual differences of HD-sEMG as the second defense can still achieve excellent performance. To improve the current state-of-the-art identification accuracy, we acquired 256-channel forearm HD-sEMG and decoded high-resolution neuromuscular information in temporal–spectral–spatial domain. We achieved a high identification accuracy of 99.85% on a 200-account (20 subjects accounts per subject) recognition task, with training and testing data acquired 3 to 25 days apart. Moreover, to address the concern of "unknown identities," we applied an "authentication + identification" validation, achieving high accuracy of 93.81% on a 200-account [(16 enrolled subjects +4 unknown subjects) accounts per subject] task. Our work substantially improves the current state-of-the-art accuracy for cross-day s- MG biometric identification (improved from % to >99% with a similar number of identified classes).
engineering, electrical & electronic,instruments & instrumentation