SGGW researchers use AI to assess milk quality
SGGW researchers have developed a method that may help milk wholesalers to determine whether raw material delivered to a dairy is free from potential fraud in a quick and efficient manner. The authors of the new solution emphasize that its application will make it possible to eliminate so-called milk fraud at the very first stage of the supply chain.
The technology has been developed at SGGW to detect fraud that may occur at the stage of raw milk delivery to dairies. The new solution, called MFA (Milk Fraud Analyzer), is based on artificial intelligence and neural networks, and was developed by the dr hab.Marcin Gołębiewski, SGGW professor, head of the Institute of Animal Science and his team.
‘The developed method automatically verifies whether the milk has been clarified to remove somatic cells from the milk before it is delivered to the dairy. This procedure usually occurs when producers want to hide the fact that the milk comes from sick cows. It is important to add that this kind of illegal procedure is barely detectable using standard analytical procedures. Our method offers a completely different approach, as we do not analyse the chemical composition but conduct an automated analysis of milk fat fraction using a microscopic image of a milk sample and a specially trained neural network’, explains prof. Gołębiewski.
What is milk fraud?
The standards required by the dairy industry indicate that there should be no more than 400,000 somatic cells per milliliter of raw milk. Somatic cells are mainly leukocytes (live and dead), i.e. the basic elements of the body’s immune system. A high concentration of leukocytes in the milk indicates that it comes from a cow that has developed an inflammation of the mammary gland and has developed a tissue infection. When one cow is ill and the milk comes from multiple individuals, the problem is reduced because the non-normative milk from the ill cow is diluted. More serious is the case when more animals or the entire herd are affected, then the number of somatic cells in the milking exceeds acceptable standards, sometimes by several times. Such raw material should not be subjected to further processing in accordance with production procedures appropriate to accepted standards for drinking milk. Some producers, being aware that the milk comes from diseased animals and wishing to sell it, try to deceive the laboratory equipment and, thus, centrifuge the milk. This causes the somatic cells to precipitate, settling on the blades and components of the centrifuge. As a result, the dairy’s laboratory will show the correct number of somatic cells on the flow cytometer and milk from sick cows will be allowed for further processing.
Milk Centrifugal Separation reduces the quality of the milk and destroys its lipid structures
The effect of centrifuging the milk is to destroy the envelope phospholipids, commonly known as the milk fat globules. These structures break down and deteriorate, which results in a lower productivity of such milk in further processing. The quality of such raw material decreases considerably. ‘The fraud based on somatic cell centrifugation not only leads to the cheating of food producers and the use of milk from sick cows, but also affects the quality of the raw material and of the products made from such centrifuged milk. Our solution effectively counteracts this, as it can uncover the fraud almost automatically’, adds dr Gołębiewski.
AI and neural networks have the answer
The detection of milk fraud based on centrifugation can be achieved by examining the fat globules contained in the raw material. However, although the test seems to be simple, up to now milk consumers need the support of specialists and the necessary equipment to carry it out, mainly the microscopes. Furthermore, there is some margin for error in this method, regardless of the use of a microscope and human assistance. The MFA system developed at SGGW improves and speeds up the whole process, using a neural network to reduce the scale of incorrect readings. ‘The Milk Fraud Analyzer replaces human eyes and speeds up analysis. The examination is reduced to the process of preparing and applying a milk sample for microscopic examination and taking a photograph of such a sample. The photograph is sent to our system, where it is subjected to automatic analysis. The analysis is based on algorithms, AI and a neural network. The MFA immediately verifies the uploaded image and sends back feedback on whether the milk has been centrifuged’, explains prof. Gołębiewski.
The efficiency of milk fraud detection with the MFA solution exceeds 85%, which, considering the initial stage of the system, should be regarded as a very high indicator. It is comparable to the results of the work of specialists. It is important to note here that such an examination can be carried out with a high frequency for practically every delivery of fresh milk. ‘Currently, the quality control tests for milk centrifugation are carried out at random, once or twice a week. By improving the analyses, milk consumers could significantly tighten up the control system and, eliminate milk fraud completely.
Moreover, it is worth mentioning that the involvement of specialists is not required to carry out analyses using our method. It is enough to prepare the sample carefully for microscopic examination and send a picture of it back to us’, says prof. Gołębiewski.
The system developed at SGGW operates both as a web service (SaaS) and in a traditional on-premise model. In the first case, the user does not have the software, but simply sends the photographs taken by the microscope over the Internet, after which a feedback is provided. In the second model of use, the software can be installed at home on a dedicated server.
Milk fraud, which involves separating somatic cells, is a major challenge for the dairy industry because of the difficulty in detecting fraud. There are no exact figures on the scale of milk centrifugation practices, but it is estimated that in Poland between 10 and 15 percent of the raw material is centrifuged before being returned to the dairy. A major advantage of the invention developed by Prof. Gołębiewski’s team is the fact that it can easily be used on an industrial scale. Due to minimal human involvement and automated processing, it can significantly contribute to eliminating the cases of milk fraud from the industry.
The MFA is also very convenient. All that is needed is to take a photograph of the sample to be tested to get a reliable result. I do believe that this solution will be of great interest to milk and dairy producers.