Abstract:

In this work, near-infrared spectroscopy (NIR) was applied for the rapid determination of 4 physicochemical
indicators of sunflower oil with the aid of several quantitative methods. The basic reference values conjugated dienes
(K232) and conjugated tienes (K270), p-anisidine value (p-AV), and total polar compounds (TPC) of the oxidized sunflower
oil samples, were obtained by chemical methods. The effects of data pre-processing method and modeling method on the
accuracy of calibration models were investigated. The results indicated that the precision of analysis for K232, K270, and
p-AV could be improved when the original spectra were pre-processed by Norris smoothing algorithms combined with first
derivative or with second derivative, while the analysis for TPC could provide the best result with the original spectra. After
spectral pretreatments, the quantitative analysis calibrations of the 4 physicochemical indicators could be developed using
partial least square regression method, and the RC were 0.98, 0.99, 0.99 and 0.98, respectively, with RMSECV of 1.37, 0.38%,
0.51 and 0.46, and RMSEP of 1.24, 0.18%, 0.40 and 0.16, respectively. The predicted values were accurate and reliable.
Therefore, NIR combined with chemometric methods could be used to determine the physicochemical indicators of edible
oils with rapidity and accuracy.

Key words: sunflower oil, physicochemical indicators, near-infrared spectroscopy, chemometry, partial least squares