Abstract: For rapid detection of the contamination level of deoxynivalenol (DON) in wheat and its products, a total of 98 samples of wheat, flour and flour products were collected and subjected to attenuated total reflectance-Fourier transform infrared spectroscopy (ATR-FTIR) analysis in the wavenumber range of 4 000 to 600 cm-1. Quantitative models between ATR-FTIR spectra and DON concentration were established by partial least squares regression analysis (PLSR) or stepwise multiple linear regression (SMLR). The results showed that the absorption values of samples with various DON contents were markedly different at 1 740, 1 648, 1 549 and 900–1 300 cm-1. Both PLSR and SMLR models could effectively predict DON contamination in samples. The coefficient of determination for prediction (RP2), root mean square error for prediction (RMSEP) and residual predictive deviation (RPD) of the PLSR model were 0.86, 0.438 mg/kg and 2.6, respectively. The SMLR model built with 9 wavenumbers was found to be optimal, with RP2, RMSEP and RPD of 0.86, 0.426 mg/kg and 2.6, respectively. These results indicated that ATR-FTIR offers the feasibility for rapid determination of wheat samples contaminated by DON.

Key words: wheat, DON, ATR-FTIR, quantitative detection, PLSR, SMLR