Investigation of Ratiometric Indices for Determining the Technological History of Dairy Systems Using Raman Spectroscopy

Determination of the history of food processing, specifically the detection of undeclared use of reconstituted ingredients instead of fresh ones, remains a serious challenge. In dairy systems, such detection is difficult because conventional chemical markers reflect the cumulative history of time and temperature changes rather than specific processing steps. Although Raman spectroscopy is widely recognized as a non-destructive, high-throughput tool for screening molecular alterations, physicochemically interpretable ratiometric indices are still required, particularly for use in explainable AI models. To identify diagnostic spectral variances, a single batch of raw whole milk was evaluated across distinct states: raw, pasteurized (76, 88, and 98 °C), and spray-dried at outlet air temperatures of ≤70 °C and ~95 °C. The resulting whole milk powders were reconstituted to a 12.1% total solids mass fraction. Raman spectra were acquired using an M532 microscope (λ = 532 nm), and the analytical band areas at 1005, 1063, 1083, and 1121 cm−1 were identified as key diagnostic markers. The structural indices Ilac and Iprot, which are mathematically orthogonal within the principal component space, were subsequently established. As a preliminary proof-of-concept, a binary logistic regression model achieved complete geometric separation between non-dried milk (including high-temperature pasteurized samples) and reconstituted milk. Within the training dataset, the false-positive rate for non-dried milk did not exceed 3.2%, whereas reconstituted milk classification model-predicted probability of belonging to the reconstituted class ranged from 96.8% to 99.9%. Furthermore, the proposed indices show great promise for classifying milk powders based on their heat treatment level. Nevertheless, further validation on independent commercial batches is strictly necessary to evaluate the proposed method’s robustness against matrix and instrumental variability.