J Food Sci. 2026 Jan;91(1):e70831. doi: 10.1111/1750-3841.70831.
ABSTRACT
Milk adulteration is a threat to food safety and public health. It is especially tricky in regions where getting quick, inexpensive tests is not always an option. To tackle this, we have come up with a simple, affordable approach. It uses easy image analysis combined with machine learning to detect common adulterants in milk. Instead of relying on laboratory equipment, all we need are straightforward optical images-just photos of milk samples. From there, we look for clues that indicate contamination, making the whole process more accessible and practical. We had taken digital photographs of milk samples mixed with water, detergent, starch, and synthetic milk. From these images, we found key statistics: the average brightness, how much it varies, and higher order features like skewness and kurtosis. These help us understand how the light scattering and turbidity change as adulterants are added. Using principal component analysis (PCA), we simplified the data, and then support vector machines (SVMs) helped classify that samples were adulterated versus pure. The results were not perfect, but they were promising. Our model hit about 85% accuracy in identifying adulteration across the different types and amounts of contaminants. In particular, adding water made the mean intensity increase, whereas detergent, starch, and synthetic milk each produced unique patterns in skewness and kurtosis due to their scattering effects. Compared to expensive spectroscopic solutions, this approach is faster, does not require chemicals, and is economical, making it ideal for quick checks, even right at the point of sale or in rural areas.
PMID:41574420 | DOI:10.1111/1750-3841.70831
