The 'electronic nose' has an accuracy rate of identifying whiskey as high as 96%. Netizens: Give one to Moutai too

Domestic Moutai and some high-end foreign whiskeys are expensive, but they are also important targets for counterfeiting.

How can ordinary people quickly determine the quality and authenticity of wine without a sommelier?

Recently, a group of engineers developed an "electronic nose" called NOS.E, which is specially used to smell wine.

It can "smell" different whiskey styles, brands and origins in less than 4 minutes, opening up new ideas for wine appreciation.

Why rely on "smell" instead of "taste"?

In fact, characteristics such as taste, smell, texture and color of whiskey can provide effective information for its evaluation.

Among them, smell is the main factor that affects the taste of wine. Researchers used this as the main breakthrough to design NOS.E.

At the 2019 Australian CEBIT trade show, they tested six whiskeys with NOS.E: among them, the regional accuracy rate was 100%, the brand name accuracy rate was 96.15%, and the style accuracy rate was 92.31 %.

Although it’s called an “electronic nose”, it doesn’t really look like a nose!

In April this year, the research paper was published in IEEE Sensors, a journal of IEEE.

After seeing this news, some netizens said excitedly: Finally, we can identify fake wine!

# Some netizens joked that one should be developed for Moutai.

So, how does this nifty and practical wine tasting tool work? Is it really reliable?

Sample pretreatment before testing

Previously at the CeBIT trade show, the developers of NOS.E tested the effect of this "electronic nose" on-site.

Before the formal test, in order to control variables and reduce the interference of irrelevant variables on the results, the researchers preprocessed the samples:

They selected 6 whiskey samples as experimental subjects: 3 3 blended malt whiskeys and 3 single malt whiskeys, and equal amounts of sample were placed in individual solid phase microextraction (SPME) vials.

Heat the samples to 30°C; and use SPME fiber to sample chlorobenzene-D5 as a reference for gas chromatography.

Then, each SPME fiber collected with chlorobenzene-D5 was placed on top of each whiskey sample (not in contact with the liquid) and left to stand for 5 minutes.

Next, these SPME fibers are put into the GC×GC-TOFMS instrument in sequence, and the collected information is processed and analyzed.

The electronic nose conducted 396 tests on 6 samples

In order to imitate the human olfactory system, the researchers equipped NOS.E with a total of 8 odor sensors.

The formal test begins——

Inject air into the SPME vial containing the sample to promote the volatile organic compounds in the wine to be discharged to the electronic nose faster.

In order to imitate the human olfactory system, researchers designed eight gas sensors for NOS.E.

The electronic nose evaluates each odor detected by the molecule and then feeds the data into a computer: the collected data is normalized, and non-parametric kernel modeling ( nonparametric kernel-based modeling) preprocessing.

The modeling process is carried out on MATLAB.

In order to reduce the impact of sensor deviation, the following normalization formula is adopted:

where, y(t) and ˆy(t) Represents the response of the sensor before and after normalization respectively.

Then, NOS.E’s system extracts 9 features from the response of the gas sensor: including the maximum first derivative of the sensor response, the minimum second derivative, the maximum second derivative, the time of input and response peak intervals etc.

After feature extraction, classify the data: randomly shuffle the data set of each whiskey, and then split it 80:20 to build a training set and a test set.

For the training set, use the ten-fold cross-validation (10-CV) method to divide it into 10 subsets to learn the whiskey classification model: 9 subsets are used for training, and the remaining 1 is used for verification.

Use linear discriminant (LD), support vector machine (SVM) and subspace discriminant (SUBD)* to train classifiers and generate a set of component classifiers for building new combined classifiers .

The final analysis results are sent to the terminal by the new classifier and presented to the user.

In order to reduce accidental errors, the researchers conducted multiple experiments on each whiskey sample; and after testing each sample 10 times, replaced it with a new one of the same kind. Whiskey to reduce the impact of alcohol evaporation on the experiment.

The researchers tested a total of 396 times before and after.

NOS.E performs well in determining the origin and style of whiskey

In order to test the accuracy of the data collected and processed by NOS.E, the researchers also used the most advanced two-dimensional gas chromatography -A whiskey sample was analyzed by time-of-flight mass spectrometry (GC×GC-ToFMS) as a control.

They tested the NOS.E test results in 3 dimensions.

The first aspect is: whether each whiskey sample can be separated from each other. The accuracy of NOS.E's on-site test results is as follows:

#The second aspect is: judging the origin of various whiskey samples, NOS.E's test results have the highest accuracy It actually reached 100%.

The third aspect is: judging the style of various whiskey samples. The accuracy of the NOS.E test results is about 82% to 94%.

About the author

Wentian Zhang, the first author of the research paper, is at Shandong First Medical University and Australia He teaches at the University of Technology Sydney; his main research directions are control engineering calculations and medical calculations.

Taoping Liu from Xi'an University of Electronic Science and Technology also participated in the development of NOS.E. He graduated with a PhD from the University of Technology, Sydney, Australia. His main research directions are control engineering calculations and medical calculations.

According to a report from the University of Technology Sydney, in addition to identifying whiskey, NOS.E may also be used to detect brandy and perfume in the future.

If this research can be promoted and applied, it may also be used for more alcoholic beverages in the near future. Consumers only need to use a small electronic product to easily determine the category and authenticity of a wine.

Well, by then, Chinese drinkers should no longer have to worry about buying fake Moutai~ (manual dog head)

Paper address: https://ieeexplore.ieee.org/ document/9701291

Reference link: [1]https://www.smithsonianmag.com/smart-news/a-new-electronic-nose-may-help-sniff-out-counterfeit-whiskey-180979931 /

[2]https://en.wikipedia.org/wiki/Gas_chromatography

The above is the detailed content of The 'electronic nose' has an accuracy rate of identifying whiskey as high as 96%. Netizens: Give one to Moutai too. For more information, please follow other related articles on the PHP Chinese website!