Disentangling Fluorescence of Endogenous Fluorescent Substances from Absorption and Scattering Effects for Quantitative Prediction for Oxidation Degree of Peanut Oils

In this study, different oxidation levels of peanut oils were prepared by heating different brands of oils at different times; the peroxide value (PV) and acid value (AV) were determined as reference values. The fluorescence intensity (F), absorption (μ_a), and reduced scattering coefficients (μ’_s) of oils were obtained by using an independently developed spectra measurement system, which was based on laser-induced fluorescence and integrated sphere techniques. Principal component analysis (PCA) was conducted on three kinds of spectra; the principal components (PCs) were extracted, and the clustering trend was analyzed. Finally, the regression models for PV and AV based on different integrations of the first five PCs of three kinds of spectra were calibrated by using different algorithms: multiple linear regression (MLR), partial least squares regression (PLSR), support vector regression (SVR), and artificial neural network (ANN). The results indicated that the optimal prediction results could be achieved by ANN based on the integration of F, μ_a, and μ’_s for PV and SVR based on the integration of F, μ_a, and μ’_s for AV, with maximum determination coefficients for validation set (R²_v) of 0.873 and 0.854, respectively, and minimum root mean square errors for validation set (RMSEV) of 2.896 meq·kg⁻¹ and 0.154 mg·g⁻¹, respectively. The proposed novel method that considers the disentangling effect of μ_a and μ’_s on fluorescence can realize robust detection for the oxidation degree of peanut oils.