Characteristic Wavelengths Optimization Of NIR Spectroscopy For Soybean Oil Acid Value

Acid value is an indicator of free fatty acid content in edible oil, it is not only one of the important indices for evaluating edible oil quality, but also the basis of computing the amount of adding alkali in alkali refining and acid eliminating, therefore it must be continuously detected in oil processing. The traditional detecting methods for acid value are almost based on chemical analysis in laboratory, which are time-consuming, and much more affected by operators, moreover, these methods will generate environmentally harmful chemicals, especially not suitable for production and control. Whereas near infrared spectroscopy analysis technology possess of many advantages, such as rapid analysis, nondestructive testing, protecting environment, it has been widely used for rapid detection of food. Because of lacking of special-purpose oil near infrared spectrum analysis instrument at home now, spectral data collected by universal spectrometer contains a large amount of useless information, that is to say, among the wave bands exist serious redundancy. With the increasing of wavelength variables, it not only increases the difficulty of the model transmission, but also the hardware cost of analytical instrument exploitation. Therefore, extracting some characteristic wavelength variables which make a greater influence on the model from the spectra by specific method can simplify the model, meanwhile improve the robustness of the model.Taking the NIR detection of soybean oil acid value for example, this paper combines some intelligent information processing methods with chemometrics, investigates the application of NIR analysis technology in rapid detection for edible oil acid value from the aspects of spectral data processing, characteristic waveband selection, and characteristic wavelength variables extraction.Firstly, first grade soybean oil is used as basic material,53 samples with different acid value are acquired by adding certain amount of oleic acid to adjust the acid value. Chemical value are measured by using standard titration, the original near infrared spectral data of samples are acquired by using Fourier transform near-infrared spectrometer. Afterwards,2 abnormal samples are removed according to the deviation between the predicted value and the actual value. In the remaining 51 samples, applying Kennard-Stone algorithm to select 41 samples as calibration set for modeling,10 samples as prediction set, and smoothing is used to denoise the original spectra for following characteristic wavebands selection.Secondly, several interval partial least squares methods are applied to extracted characteristic wavebands of acid value. The characteristic wavebands ranging from 4540 cm-1 to 5346 cm-1 and 6807 cm-1 to 7004 cm-1 in which containing 262 wavelengths in total are extracted from the spectral data, the coefficient of determination R2 and the root mean square error of prediction set (RMSEP) of the regression model based on characteristic wavebands are 0.9573 and 0.1102, respectively, RSD is 6.56%, while model established by full spectra in which containing 2075 wavelengths, the R2 and RMSEP are 0.8506 and 0.2092,respectively, RSD is 12.45%, the model prediction accuracy has improved, meanwhile the complexity of the model is greatly reduced. On this basis, further characteristic wavelength variables are extracted.At last, genetic algorithm and the successive projection algorithm are proposed for characteristic wavelengths selection. As a result,29 characteristic wavelengths are selected by genetic algorithm,11 characteristic wavelengths are selected by the successive projection algorithm, after combining the selection results of these two methods,10 characteristic variables that can mostly reflect the absorption feature of oil acid value in near infrared range are picked out,using these 10 wavelengths to build a MLR regression model, The R2,RMSEP and RSD of the MLR model are 0.9890,0.0957and 5.69% when predicting the unknown samples. The research proves that doing characteristic wavelength selection in near infrared spectra of oil acid value can effectively remove redundant information, reduce complexity of the model, improve the prediction precision, and also lay the foundation of the development of special oil near infrared spectrum analysis instrument and realize on-line monitoring.