Optimization of quantitative detection model for benzoic acid in wheat flour based on CARS variable selection and THz spectroscopy

Abstract

Food safety has been an issue of common concern in today’s society. Excessive benzoic acid (BA) additives in wheat flour would pose a huge threat to human health. In this paper, terahertz (THz) time-domain spectroscopy was used to detect BA quantitatively, which is a preservative additive in flour. Samples were prepared by grinding, drying, weighing, mixing and performing successively according to the designed concentration gradient (concentration 0.04%, 0.08%, 0.1%, 0.2%, 0.4%, 0.5%, 1%, 1.5%, …, 10%), with the BA concentration ranging from 0.040% to 19.99%. Then the THz Spectra of wheat flour samples with different concentrations of BA were collected. After preprocessing the original terahertz spectra with multiple scattering correction and other methods, Competitive Adaptive Reweighted Sampling (CARS), Genetic Algorithm (GA), Principal Component Analysis (PCA) and Uninformative Variable Elimination (UVE) were used to select effective spectral bands, and then Partial Least Squares (PLS) and Least Square Support Vector Machine (LS-SVM) models were established respectively. It turned out that the absorption peak of BA was at 1.94 THz of terahertz absorption coefficient, which increased with the rise of BA concentration. The results also indicated that better modeling results could be obtained by using CARS method for variable selection. And the established CARS-PCA-LS-SVM model had the best prediction result by combining PCA with CARS, with the correlation coefficient of the prediction (Rp) of 0.9956 and Root Mean Square Error of Prediction (RMSEP) of 0.64%. Hence, it can be concluded that terahertz technology combined with LS-SVM model can well achieve the quantitative detection of BA concentration in flour. Although the model is not universal since samples size is not large enough, the research method provides a basis for the detection of BA additives in flour.