Combined with machine learning, this 'artificial nose,' built with a 16-channel sensor array, was able to authenticate up to 20 individuals with an average accuracy of over 97 per cent.
Researchers from Kyushu University's Institute for Materials Chemistry and Engineering, in collaboration with the University of Tokyo detail their findings in Chemical Communications.
There are a variety of biometrics that machines can use to identify people, including fingerprints, palm prints, voices, and faces to the less common options of ear acoustics and finger veins.
"These techniques rely on the physical uniqueness of each individual, but they are not fool proof. Physical characteristics can be copied, or even compromised by injury," said Chaiyanut Jirayupat, first author of the study. "Recently, human scent has been emerging as a new class of biometric authentication, essentially using your unique chemical composition to confirm who you are."
One such target has been percutaneous gas, which are compounds produced from skin. These methods have their limits because the skin does not produce a high enough concentration of volatile compounds for machines to detect.
"The concentration of volatile compounds from the skin can be as low as several parts-per-billion or trillion, while compounds exhaled from the breath can go as high as parts-per-million," said Jirayupat. "In fact, human breath has already been used to identify if a person has cancer, diabetes, and even COVID-19."
The team began by analysing the breath of subjects to see which compounds could be used for biometric authentication, and 28 were found to be viable.
Based on this, they developed an olfactory sensor array with 16 channels, each which could identify a specific range of compounds. The sensor data was then passed into a machine learning system to analyse the composition of each person's breath and develop a profile to be used to distinguish an individual.
Testing the system with breath samples from six people, the researchers found it could identify individuals with an average accuracy of 97.8 per cent. This high level of accuracy remained consistent even when the sample size was increased to 20 people.
"This was a diverse group of individuals of differing age, sex, and nationality. It's encouraging to see such a high accuracy across the board," said Takeshi Yanagida who led the study.
He added that more work is needed before it can be used to unlock a smartphone.
"In this work, we required our subjects to fast six hours before testing," said Yanagida. "We've developed a good foundation. The next step will be to refine this technique to work regardless of diet. Thankfully, our current study showed that adding more sensors and collecting more data can overcome this obstacle."