2022年第83回応用物理学会秋季学術講演会

講演情報

一般セッション(口頭講演)

12 有機分子・バイオエレクトロニクス » 12.7 医用工学・バイオチップ

[22p-A105-1~20] 12.7 医用工学・バイオチップ

2022年9月22日(木) 13:00 〜 18:45 A105 (A105)

宮本 浩一郎(東北大)、笹川 清隆(奈良先端大)、山下 一郎(阪大)、野田 実(京工繊)

17:30 〜 17:45

[22p-A105-16] Breath odor-based Individual Authentication by An Artificial Olfactory Sensor System and Machine Learning

〇(P)Chaiyanut jirayupat Jirayupat1、Kazuki Nagashima1,2、Takuro Hosomi1,2、Tsunaki Takahashi1,2、Wataru Tanaka1、Masaki Kanai3、Takeshi Yanagida1,3 (1.The Univ. of Tokyo、2.JST PRESTO、3.IMCE, Kyushu Univ.)

キーワード:Breath sensing, Biometric autentication, Artificial Olfactory System

In the digital age, an encounter with cyber thieves is a serious problem so biometric authentication has become a frontier technology for increasing the security levels of digital privacy. Although physical information-based techniques based on fingerprint and face recognitions are mainly utilized for biometric authentication, the competition between the biometric authentication techniques and ones for hacking them is rapidly growing neck and neck. Human scent analysis/sensing is a new class of biometric authentication techniques using chemical information. Since human scents such as exhaled breath and percutaneous gas have a strong genetic basis, their chemical composition profiles are inherently different among individuals and therefore can potentially be utilized for individual authentication. Among human scents, exhaled breath is known to have thousands of volatile organic compounds (VOCs) and allows us a facile and non-invasive sampling. Moreover, the breath odor is consumed once it is utilized, which may reduce a risk of long-term presence attack. Thus, the breath odor sensing has great potential to realize a secure individual authentication.
In this research, we aim to demonstrate the feasibility of breath odor sensing-based individual authentication using an artificial olfactory sensor system. A 16-channel chemiresistive sensor array was utilized for detecting various VOCs contained in breath odor. The acquired sensing responses from 20 persons under fasting condition were analyzed by machine learning with a neural network algorithm and the identification accuracy of individual authentication was evaluated. The result shows a high mean accuracy of over 97% for classification and highlights the impact of the number of sensors on the accuracy and reproducibility.