Abstract: The viability of using near-infrared (NIR) spectroscopy to detect illegally added kojic acid (KA) as a browning inhibitor in wheat flour was studied. For this purpose, three common types of commercial flour, i.e., high-gluten flour, plain flour and low-gluten flour, were added with different amounts of KA (0.0%, 0.5%, 1.0%, 3.0%, 5.0%, and 10.0%), respectively. NIR spectra of all samples were collected in the wavelength range of 1 000–2 400 nm. For high-gluten flour samples, three common spectral preprocessing methods were compared with each other, as well as with non-preprocessing using partial least squares (PLS) regression. Savitzky-Golay derivative (SGD) was found to be the best preprocessing method. Then interval partial least squares (iPLS) was adopted to obtain optimized spectral interval in the wavelength range from 1 088.8 to 1 153.5 nm. The PLS model based on the optimal spectral interval showed better performance than that in the full wavelength range. Moreover, a PLS model was developed based on the optimal spectral interval for plain flour, low-gluten flour and a mixture of all three types, respectively. The models for all three types of flour and their mixture showed a determination coefficient (R2) of 0.949–0.972, a root mean square error (RMSE) of 0.581%–0.830%, and a ratio of standard deviation of the validation set to standard error of prediction (RPD) of 4.171–4.830. The model exhibited a good prediction performance for wheat samples with KA contents of 1.0%–10.0%. These results indicated that NIR spectroscopy could be used as an auxiliary method for precise chemical detection to improve the detection efficiency for massive samples.

Key words: kojic acid, wheat flour, near-infrared (NIR) spectroscopy, partial least squares (PLS), interval partial least squares (iPLS)