Rapid Determination of Benzo(a)pyrene Concentration in Soybean Oil by Terahertz Transmission Spectroscopy with Chemometrics

Benzo(a)pyrene (B(a)P), as the best known of the group I carcinogen, is unavoidably generated in production process of soybean oil. The feasibility of rapid determination of B(a)P in soybean oil with terahertz spectroscopy was examined. Sparse representation (SR), as a pre-treatment method, was used to solve the problem of the redundant information of THz signal. Different chemometrics combined with THz signal were compared to find the best determination model. The results showed that SR-LS-SVM was the best method with the accuracy of 96.19% in B(a)P concentration discrimination experiment. And it was feasible to detect the limit concentration of B(a)P over 2 µg/kg in soybean oil (over 90% accuracy). In quantitative detection experiment, the correlation coefficient of prediction (RP) was 0.9623 and the root-mean-square error of prediction (RMSEP) was 1.2836 µg/kg. It could be concluded that THz combined with chemometrics would be an attractive technique for rapid determination of the B(a)P concentration in soybean oil.