Research On Detecting Methods For Soybean Oil Colour Based On SVM And Near-infrared Spectrum Analysis

Edible oil is an important part of the human diet and one of the three major nutrients of the human. Its quality situation will affect the development of food industry, health security of consumers and social harmony and stability. The depth of colour is one of the important indices for vegetable edible oil quality, particularly for edible vegetable oil, and often requires having a lighter colour. Therefore, you can understand the pure degree of edible oil, refined level as well as to determine whether deterioration by determining of edible oil colour, it is has important significance to improve the conditions of the oil processing and evaluation of oil quality. The traditional detecting methods are all based on chemical analysis in laboratory, much more affected by operators, low determination speed, more chemical reagent and the complex sample pretreatment. Using near infrared spectral analysis technology to detect Soybean oil colour rapidly can overcome the drawbacks of conventional methods, easy realize automation and possesses of realistic significance to improve product quality of edible oil, realize dynamic monitoring oil processing process.This paper took the soybean oil colour for research object, combined chemometrics methods, under environment of MATLAB, discussed the application of near infrared spectroscopy analysis in rapidly quantitative detection of edible oil colour from the aspects of spectrum data preprocessing, and constructing correcting model and so on.First the three different level oils with different Lovibond yellow values are classified by C-SVM, compared to the original spectrum and wavelet spectrum, and compared to the selection of kernel function and nuclear parameters optimization. Finally the SVM classifier suitable for near infrared spectral recognition of oil colour is designed, recognition correct rate of different level oils achieved100%. On this basis, the regression between oil near infrared spectral data and Lovibond red values with different yellow values is made by using PLS and ε-SVM, modeling and forecasting analysis is made by the spectrum through wavelet smoothing-normalized-baseline-SNV. The results show that, SVM modeling method has more advantages compared to traditional PLS modeling method. The decision coefficient R2of Yellow20, Yellow35, Yellow70achieved0.999516,0.91037å'Œ0.9104.The decision coefficient RMSEP of oil acid and peroxide value achieved0.123917%,,1.47954%å'Œ1.47965%, exceeded the expected goal.The research shows that realizing qualitative and quantitative analysis of oil colour using near infrared spectral technology is feasible, also established foundation for further realizing real time on-line detection and control of oil colour.