Quantitative and qualitative analysis of buffalo milk adulteration using Raman spectroscopy and chemometric–deep learning models

Abstract

Buffalo milk is nutritionally rich but vulnerable to adulteration, posing challenges to food safety. This study analyzes buffalo milk, soybean milk, Holstein milk, and five common additives—ammonium chloride, urea, sodium bicarbonate, sodium citrate, and sucrose—using Raman spectroscopy. Six spectral preprocessing methods were systematically evaluated to enhance model performance. For qualitative detection, PLS-DA models combined with Multiplicative Scatter Correction (MSC) preprocessing achieved excellent classification accuracy (up to 100 %) for pure buffalo milk, water, and soybean milk adulteration.

For quantitative analysis, both PLS and MSC-CNN regression models were developed. The MSC-CNN model achieved high predictive performance for sodium bicarbonate (R² = 0.97) and sodium citrate (R² = 0.93), with RMSEP <5 % of full scale. Detection limits were as low as 17.4 mg/kg for sodium bicarbonate and 20.9 mg/kg for sodium citrate, meeting practical sensitivity requirements. Compared with existing PLS-based methods, our approach improved predictive accuracy and expanded low-concentration detection. The proposed Raman–deep learning strategy offers a rapid, accurate, and non-destructive solution for milk adulteration monitoring and quality control.