Rapid detection and spectroscopic feature analysis of mineral content in camel milk using fourier-transform mid-infrared spectroscopy and traditional machine learning algorithms

Abstract

Camel milk is rich in nutrients and bioactive factors, with mineral content generally higher than that of cow milk, but there is currently no internationally established, rapid, batch-testing method for the mineral content. This study collected samples of camel milk from 113 locations in Xinjiang, China. For the first time internationally, based on the true mineral values determined by ICP-OES (Inductively Coupled Plasma Optical Emission Spectroscopy) and the extracted mid-infrared spectra data, a quantitative prediction model for 10 key minerals (Ca, Fe, K, Mg, Mn, Na, P, Sr, Zn, and Se) was established using Fourier-Transform Mid-Infrared Spectroscopy (FT-MIRS) and the traditional machine learning algorithm Partial Least Squares Regression. The R_t² of the test set ranged from 0.61 to 0.91, RMSE_t ranged from 2.21ug/kg(Se) to 197.08 mg/kg(K) and the RPD_t from 1.59 to 3.28. In addition, the distribution, patterns, and correlations of mineral-related characteristic wavenumbers in camel milk were summarized. This study opens a new avenue for the rapid detection of minerals in camel milk and fills the research gap in in using FT-MIRS to detect mineral content in camel milk.