Researchers in China have pioneered a rapid, green, and non-destructive detection system using NIR spectroscopy and machine learning to ensure yak milk powder quality.
Recently, a collaborative study conducted by researchers from Kunming University of Science and Technology, Nankai University, and the Kunming Institute of Food and Drug Control evaluated a new, rapid, and inexpensive method for assessing the quality of yak milk powder. This study, published in Food Control, showcased a new miniaturized near-infrared (NIR) spectroscopy method coupled with chemometric modeling to measure the protein, fat, moisture content, and acidity in yak milk powder (1). Their findings suggest that this method can help improve the detection of yak milk powder while solving the challenges of traditional methods.
Yak milk is an alternative to cow milk that is mostly consumed by people who live in the highlands where access to cows is limited or nonexistent. Yak milk differs from cow milk because it has higher fat and protein content (2). Yak milk also contains more calcium and iron than typical cow milk (2). Traditionally, determining the quality of yak milk powder has relied on time-consuming and labor-intensive methods involving chemical reagents, such as the Kjeldahl method for protein, Soxhlet extraction for fat, and oven-drying for moisture. These techniques, though accurate, are impractical for large-scale or real-time quality monitoring in commercial production environments (1).
Recently, spectroscopic techniques have been used to assess the quality of milk and other dairy products (3,4). In this study, the researchers tested another novel method that utilized miniaturized NIR spectroscopy with machine learning (ML) algorithms. NIR spectroscopy is a non-destructive technique that requires no chemical reagents (1). By using this method, the researchers demonstrated how it can analyze yak milk powder in seconds while maintaining accuracy.
As part of the experimental procedure, the researchers collected 100 samples of yak milk powder from nine different brands across four major yak-rearing provinces in China, which included Sichuan, Gansu, Qinghai, and Yunnan (1). These regions allowed for sample variability, which helped make the data set highly representative of the yak milk powder market (1).
Then, the spectral data was preprocessed. The researchers used Savitzky–Golay (SG) smoothing to reduce noise and enhance signal quality. The researchers then applied three machine learning models to the processed data: partial least squares combined with support vector regression (PLS–SVR), ridge regression (RR), and random forest (RF) (1). The goal was to identify which model provided the most accurate predictions of the four quality indicators.
Among the models tested, the random forest (RF) algorithm demonstrated the highest prediction accuracy. The correlation coefficients for protein, fat, and moisture content were 0.9846, 0.9642, and 0.9915, respectively, whereas the acidity prediction achieved a correlation coefficient of 0.9819 (1). The model also showed strong predictive stability, with RPD (residual predictive deviation) values ranging from 5.4 to 20.1, indicating excellent model performance (1).
Furthermore, the model's performance makes it suitable for integration into real-time monitoring systems, providing a new pathway for automation in dairy processing facilities. The researchers suggest that future research endeavors could expand this system's scalability and reliability by regularly updating the training model with fresh sample data, as external factors like seasonal variation, processing techniques, and storage conditions could influence prediction accuracy (1).
Compared with traditional methods, the new approach not only shortens detection time from hours to minutes but also aligns with the broader global push for sustainable and reagent-free analytical technologies. The research shows how miniaturized NIR spectroscopy can significantly reduce analysis time, eliminate the use of harmful reagents, and improve operational efficiency in production lines (1)
As demand grows for healthier, traceable, and efficiently produced food products, such innovations could help ensure quality and safety while reducing the environmental impact of producing and testing milk products.
