Identification And Analysis Of Safflower Seed Oil Adulteration Based On Visible-near Infrared Hyperspectral Technology

In this study,visible-near-infrared hyperspectral combined with machine learning was used to qualitatively and quantitatively analyze the adulteration of safflower seed oil,and to explore the correlation between spectral information and chemical composition in the sample.The main research content includes the following parts:（1）taking safflower seed oil,sunflower seed oil,corn oil,and soybean oil as the research objectives,five kinds of mixed oil samples and four kinds of pure vegetable oils were qualitatively identified by visible-near infrared hyperspectroscopy combined with machine learning.The average spectral data of 19 kinds of oil samples were collected by visible-near infrared hyperspectroscopy,and four preprocessing methods were used to preprocess the spectral data to construct the data set.A variety of machine learning models were established to qualitatively identify many kinds of oil samples.Finally,the combination of MF-GBDT-GBDT is obtained,and the fast recognition rate of four kinds of oil samples is achieved.（2）the content of safflower seed oil in different oil samples was quantitatively analyzed by visible-near infrared hyperspectroscopy combined with machine learning.The spectral data of oil samples were collected by the visible-near infrared spectrometer,and the spectral data sets with 7 kinds of volume fractions were constructed.A variety of machine learning regression analysis models were established to identify the content of safflower oil in different sample oils.Get the best regression model MF-Ridge-Stacking（using Ridge and PLSR as the basic model and Light GBM as the metamodel）.When the number of input features is greater than or equal to 46,the test set R² is stable above 0.95.the R_cv² is maintained around 0.93.the model results are stable.（3）the contents and concentrations of linoleic acid,oleic acid,and palmitoleic acid in19 kinds of oil samples were determined by gas chromatography-mass spectrometry.To explore the internal relationship between the contents of three fatty acids and spectral data.The results show that there is a strong correlation between doping concentration and linoleic acid content,and the best model MF-Ridge-Stacking（using Light GBM and GBDT as the basic model and Cat Boost as metamodel）is obtained by quantitative analysis combined with machine learning.When the number of input features is 40,the test set R²is 0.87 and R_cv² is 0.664.The results show that visible-near infrared hyperspectroscopy combined with machine learning can be used for rapid qualitative and quantitative analysis of adulterated safflower seed oil.