"Electronic nose" and computer vision will help determine the readiness of dishes

“Electronic nose” and computer vision will help determine the readiness of dishes


Ekaterina Alexandrova | 01/14/2021

Skoltech researchers have figured out how to use chemical sensors and computer vision to determine if a dish has been cooked correctly.

This method can be used by restaurant chefs to control and automate food preparation processes. The developers do not exclude that such functions may appear in home “smart”

Professor of Skoltech and Aalto University Albert Nasibulin, senior researcher at Skoltech Fyodor Fedorov and their colleagues have created an “electronic nose”, which is a set of sensors, using
which you can register the “profile” of the smell in order to “sniff” the chicken during the frying process, and also applied computer vision algorithms to “look” at the chicken. “Electronic nose” is simpler and
cheaper to operate than, for example, an expensive gas chromatograph or mass spectrometer. In previous works, other researchers have already shown that the “electronic nose” is able to distinguish between cheeses
by varieties and reject rotten apples or bananas, and computer vision allows us to recognize visual structures, for example, cracked biscuits.

The Skoltech Nanomaterials Laboratory, headed by Professor Nasibulin, is developing new materials for chemical sensors that can be used, in particular, in the HoReCa segment.
to control the quality of air filtration in ventilation systems. One of the authors of the article, Skoltech Master’s student Aynul Yakin underwent a summer industrial practice in Novosibirsk, where
tested the efficiency of industrial filters from a large Russian manufacturer using laboratory sensors. The results of this project became the basis for the experiment on
drawing up a “map” of grilled chicken smells.

“It is not enough to rely only on the electronic nose to determine culinary readiness; it is necessary to use computer vision in addition to it. Both of these tools combined
form the so-called “group of electronic experts”. Our work was attended by CDISE specialists with extensive experience in the field of computer vision. Together we checked and confirmed
the hypothesis that the combined use of computer vision and the “electronic nose” allows more precise control over the cooking process “, – said Albert Nasibulin.

The researchers combined the two methods to create an accurate, non-invasive method for determining the degree of cooked food. For this purpose, they chose chicken meat that is popular all over the world and purchased from
one of the Moscow supermarkets a batch of chicken breasts. In the process of preparing the grilled chicken, the researchers trained their tools to evaluate and predict the degree of cookiness of the dish.

Researchers have developed their own design for an “electronic nose” consisting of eight sensors to detect smoke, alcohol, carbon monoxide and other compounds, as well as temperature and
humidity. Then the “electronic nose” was placed in a ventilation system. In the process of frying, the chicken was photographed, and the information obtained was fed to the input of algorithms that
classification or search for patterns in the data. Changes in environmental composition at different stages of the frying process were also recorded using differential mobility analysis for
measurement of the size of aerosol particles and the method of mass spectrometry.

The most important part of the experiment was a “tasting” with 16 graduate students and researchers who were asked to rate tenderness, juiciness, richness, appearance and overall roast.
chicken breast on a 10-point scale. The feedback received was then compared with the results of the analysis to check the consistency between the analytical data and the results of subjective perception.

The researchers roasted chicken on the grill right outside their laboratory, and used the Skoltech canteen as a test site. “In the context of the coronavirus pandemic, we had to work in masks and
test in small groups; it was not entirely customary. For the correctness of the experiment, we carefully instructed all the participants and gave each of them a testing protocol. we
prepared a large number of samples, assigned a code to each, and participants tested the samples blindly. For all participants involved mainly in materials science and using complex
analytical methods, it was very interesting, because chicken meat is also a material! ”, – Fedor Fedorov notes.

The researchers note that their system successfully coped with the task of recognizing the main degrees of roast (“undercooked”, “overcooked”, “overcooked”), therefore, in principle, it can be used for
purposes of automation of food quality control in food production. The authors also note that for cooking other parts of the chicken, such as legs or wings, or using other
preparation method will require retraining of electronic “eyes” and “nose” on the new data.

Scientists plan to test their sensors in real conditions in restaurant kitchens. Another possible application of the new sensor could be the detection of spoiled meat at the stage when
the human sense of smell is not yet able to detect changes in its smell.

The research was carried out with the support of the Russian Science Foundation (RSF). An article detailing the results of the study was published in the scientific journal Food Chemistry.

Machine vision

Skolkovo Technopark

Leave a Comment