Studies On Discrimination Of Honey Adulteration By Near Infrared And Raman Spectroscopy

Near infrared spectroscopy is efficient for the detection of reducing sugar in honey, which cost little time and be with high accuracy. Meanwhile, data pretreatment methods, such as MSC, Norris smooth, first derivative, second derivative, are useful treatment to improve the accuracy of the model. The prediction accuracy of the model is higher than those built by overseas researchers.This article studied the test methods of fructose, glucose, sucrose and maltose by Near-infrared spectroscopy. The results show that, Kennard-Stone method is an effective algorithm to devise the calibration set and the validation set. Monte Carlo cross validation was used to exclude the outliers. The determination coefficients of predict value and the real value fructose, glucose, sucrose and maltose are 0.945,0.927, 0.901,0.610 and 0.971,0.940,0.949,0.750 separately before and after excluded the outlier. It efficiently improved the data distribution and the accuracy of the model. The model built with Near infrared spectroscopy after the Wavelet transform is no significant change, but it is less time consuming, which is 2.76s other than 3.65s before the Wavelet transform.The honey samples that were used to build model, were collected from teens of provident. the geographic and pollen of the honey have some impacts on the spectrum. It is wider implicated under complex conditions. To adapt the need of the actual detection, the adulteration samples were included in the model.By using the chmometric methods, the results predicted by the Near infrared spectroscopy are accurate and reliable. So it is feasibility to build a rapid online testing technology for the quality control of honey. The cross validation coefficient of determination of HMF and total acid were 0.571 and 0.640 separately.The best prediction accuracy of the BP neural network is over 90% in all the different ways of adulteration in honey samples including adulterate sucrose, glucose separately and adulterate sucrose and glucose together. But the BP neural network is with poor stability, and it had got totally different discriminate results for the same set of data. Using RBF neural network instead,47 iterations was chosen, and set the MSE as 0.05, the error rate of validation set and test set were lower than that of BP neural network, and the test set error rate is O. Moreover, RBF neural network model is with a good repeatability.In summary, the near-infrared spectroscopy meet the demand of high performance testing and quality control technology for reducing sugar, fructose, glucose, sucrose and maltose.The feasibility of using Raman spectroscopy to determine chemical composition of honey was examined. The influences of spectrum pretreatment on the performance of calibration equations were also studied. The regression model of Support Vector Machines (SVM) was selected for the calibration of honey fructose, glucose, sucrose and moisture. Good predictions were obtained for fructose, glucose and moisture with squared correlation (R2) of 0.84,0.87 and 0.75 between the predicted values and the reference values. The prediction accuracy for sucrose in honey was poor and unreliable. The study indicates that Raman spectroscopy can be used for fast determination of major components in honey.In the qualitative identification of adulteration in honey, the baseline correction is unnecessary. After the centralization of the spectrum, the PLS-LDA models had been built and the cross validation error rate were all lower than 0.04, and the correct rate of the prediction set was 100%. The less adulated volume and the more complex method of adulteration, the higher error rate.In summary, Raman spectroscopy could analyze fructose and glucose in honey quickly. However, further research in instruments and chmometrics is needed for the analysis of sucrose and maltose.