Application Of Near-infrared Spectroscopy Analysis In Quality Detection Of Edible Vegetable Oil

In combination with the chemometrical methods, the near-infrared spectroscopy analysis technology was used for edible vegetable oil quantitative analysis, in which the fatty acids (palm acid, stearic acid, oleic acid, linoleic acid and linolenic acid) were inspected. In order to provide new reference for the nondestructive examination for the edible vegetable oil quality detection, several essential issues on near-infrared spectroscopy technology in the quantitative analysis were investigated in terms of the selection of system parameters, the elimination of outlier samples and the building of calibration model. In order to evaluate the prediction accuracy rate of the correction model, the near-infrared spectroscopy analysis technology was also used to identify the different extraction techniques and different types of the edible vegetable oil and explored several key issues on the qualitative detection of near-infrared spectroscopy. The main contents and conclusions are as follows:312 edible vegetable oil samples of different brands and kinds were collected for this study, chemical value of main fatty acids consisting of palm acid, stearic acid, oleic acid, linoleic acid and linolenic acid were detected, and the near-infrared spectrum was synchronously detected by WQF-400N fourier near-infrared spectroscopy. According to the contrastive experiments, the system parameter of instrument such as the scan time, differentiate rate and sample measure time were confirmed. Applicating three methods of eliminate outlier sample, the method of mahalanobis distance can comfirm the outlier spectrum. The result showed that this method was more suitable for this study. After 6 outlier spectrums eliminated, the spectrum of 306 samples were ensured as calibration set and prediction set. The calibration and predicition samples were differentiated with cluster analysis. The numbers of calibration samples and predicition samples were 203 and 103 respectively. AKS algorithm was used in calibration samples for further selection, and the spectrum of 196 samples were selected as the best sample sets, while the spectrum of the rest of 7 samples were eliminated. The forecast veracity was enhanced, and it showed that optimal sample set was more representative. Comparing the effect of 6 pretreatment methods, such as standardization, the best pretreatment methods of five fatty acids were found out respectively.Meanwhile,7 optimize wavelength methods, such as Uninformative Variable Elimination by PLS and Genetic Algorithm, were compared in the optimal pretreatment method, respectively building the PLS forecast model of five fatty acids. The result showed that the effect by using Uninformative Variable Eliminationwas best.This study also made the research of NIR spectral analysis in distinguishing the manufacture technology of colza oil, peanut oil and sesame oil. In combination with ECVA, different pretreatment methods were used for building models. The forecast correct rates of 3 edible vegetable oils were all reached 100%. The result indicated it was feasible to classify 3 edible vegetable oils'manufacture technology combining with the near-infrared spectrum and suitable pattern recognition method.This article conducted the research into the recognition of edible vegetable oil's kinds in the method of near-infrared spectral analysis. The model was built respectively through Partial Least Squares Analysis combining with K-Nearest Neighbors and Principal Component Analysis combining with Least Square Support Vector Machine method through confirmation of the forecast result. The correct rates reached 95% and 97.5% respectively. The result showed that it was feasible to classify 3 edible vegetable oils' kind combining with the near-infrared spectrum and suitable pattern recognition method.In addition to this study, there were some researches to distinguish edible vegetable oil mixed with bad oil studied in the method of near-infrared spectral analysis. The edible vegetable oil sample which contained 5% bad oil in the prediction set can be accurately identified through Partial Least Squares Discriminant Analysis model. The result adequately showed that it was feasible to identify peanut oil and sesame oil mixed with bad oil combining with the near-infrared spectrum and suitable pattern recognition method.