An effective method for identifying cow milk powder adulteration levels in goat milk powder using hyperspectral imaging

ZT Cai and J Sun and L Shi and Y Liu and XH Wu and CX Dai, JOURNAL OF FOOD MEASUREMENT AND CHARACTERIZATION, 19, 8854-8868 (2025).

DOI: 10.1007/s11694-025-03576-0

Some producers adulterate goat milk powder with cow milk powder for excessive profits, which reduces the nutritional value and increases the allergy risk of their products. To tackle the prevalent issue of adulterated goat milk powder in the market, we propose a rapid and non- destructive detection method for identifying such adulteration in goat milk powder. In this study, a total of 880 samples of goat milk powder adulteration were collected using a visible-near infrared (VIS-NIR) hyperspectral imaging (HSI) system operating within the wavelength range of 397.91 to 986.02 nm. Subsequently, detrending (DT) preprocessing was performed to mitigate the influence of noise, thereby enhancing the overall quality of the data. Then, the LassoNet algorithm was introduced to extract feature wavelengths from the full wavelength range to reduce data redundancy. To address the challenge of parameter tuning, the beluga whale optimization (BWO) algorithm was introduced as a means to globally optimize the hyperparameters of the support vector machine (SVM). A classification model was developed to detect goat milk powder adulteration and assess adulteration levels using the DT-LassoNet-BWO- SVM approach. The classification accuracy achieved by this model on the training set was 95.76%, while the accuracy on the test set reached 94.55%. The results show that compared with conventional feature selection methods, the LassoNet algorithm demonstrates greater efficacy in identifying representative spectral wavelengths. Meanwhile, the BWO- SVM model demonstrates significant potential for classifying adulteration levels in goat milk powder. In conclusion, HSI technology offers a viable approach for detecting goat milk powder adulteration.

Return to Publications page