Research On Rapid Quantitative Detection Of Multi-component Blended Oil By Near Infrared Spectroscop

Edible blend oil is popular among consumers because of the nutritional balance of fatty acids.However,since the price of different pure vegetable oils varies considerably,fraudulent practices,like mislabeling the proportion of single component oil content in blends for financial benefit,have existed in the blend oil market.Therefore,it is important to develop a rapid and effective method for the quantitative determination of single component oils in edible blend oil for proportion authenticity determination and quality control.NIR spectroscopy is expected to be an effective method for the quantitative determination of edible blend oil because of its advantages of rapidity,non-destructiveness,and no sample pretreatment.However,most previous studies on the quantitative determination of edible blend oil based on NIR spectroscopy have focused on binary and ternary edible blend oil,and few studies have been conducted on higher order edible blend oils.In this thesis,the feasibility of NIR spectroscopy combined with chemometrics for the determination of single component oil content in binary to hexanary edible blend oil was investigated.The specific studies are as follows:1.Sesame oil,soybean oil,rice oil,sunflower oil and peanut oil were mixed successively with corn oil to prepare binary,ternary,quaternary,quinary and hexanary edible blend oil samples,and their NIR spectra were measured in a transmittance mode in the range of 12000-4000 cm^-1.2.Five multivariate calibration methods,including principal component regression,partial least squares regression（PLS）,support vector regression,artificial neural network and extreme learning machine,were used to develop quantitative models of single component oils in binary and ternary edible blend oil to compare the modeling effects of different methods.The PLS model was found to be superior in terms of the degree of fit,operation time and prediction accuracy.3.The effects of six spectral preprocessing methods,including SG smoothing,first derivative,second derivative,standard normal variate,multiplicative scatter correction,continuous wavelet transform and their combinations on the performance of PLS models for quaternary edible blend oil were investigated.Compared with the PLS model,the prediction accuracy of the best spectral preprocessing-PLS model for each single component oil was significantly improved.4.Four variable selection methods,including uninformative variable elimination,Monte Carlo uninformative variable elimination,random test and competitive adaptive reweighting sampling,were used to further optimize the best spectral preprocessing-PLS models for quaternary edible blend oil.After the optimization of the best variable selection method,the model is not only simplified but also the prediction accuracy is further improved.5.The optimal spectral preprocessing-variable selection-PLS quantitative models for single component oils in binary to hexanary edible blend oil were developed and used to predict the contents of the prediction set samples,respectively.The developed optimal models for single component oils in binary to hexanary edible blend oil were stable and reliable with good prediction accuracy.Therefore,NIR spectroscopy combined with chemometrics is feasible for the accurate quantitative determination of single component oil content in binary to quinary edible blend oil.In contrast,the final optimal models for sunflower oil and peanut oil were poorly predicted for hexanary edible blend oil,and the feasibility still need to be explored by other chemometrics methods.