Study On Quality Evaluation For Honey By Near Infrared Spectroscopy

With the improvement of people's living standards and enhanced awareness of health, honey, as a kind of health product with high nutritional value, is very popular for consumers and the sales are growing continuously. However, the quality of honey is not satisfied in recent years, especially in the floral origin, identification and adulteration. Therefore, it is one of most important issues in the development of apiculture to identify and regulate the honey product quality and more effective techniques are required.In the study, Fourier transform near infrared (FT-NIR) spectroscopy together with various chemometrics methods was applied to determine honey quality such as floral origin, components and authenticity. And the corresponding math models were also developed in order to evaluate quality of the honey in China.The major results and conclusions are summaried as follows:1,The spectra collection parameters on FT-NIR were analyzed. Through the comparison analysis, the optimized parameters were chosen including fiber optic diffuse transmission, 8 cm^-1 of scan resolution, 32 of scan numbers, collection temperature(40℃)and optic lengths.2,The influence of NIR spectra was analyzed based on different factors, i.e., floral origin, temperature and geographic origin. The results showed that there were significant differences between the spectra of different honey types in most of the wave bands, the spectra characteristics were obviously influenced by different temperatures and there was difference among spectra of honey samples from different geographic origins.3,Three models for discrimination of botanical origin of honey were developed by Mahalanobis distance -discriminant analysis (MD-DA), Discriminant partial least squares analysis(DPLS)and Artificial neural net-work(ANN)together with different wavelength ranges and different spectra preprocessing methods, and the performance of the models was compared. The results showed that the best discrimination models for honey types developed by MD-DA together first derivative and S-G smoothing, with correct classification of the calibration and validation sets were 87.4 % and 85.3 %, respectively. DPLS together with first derivative and 10000-4200 cm^-1 wavelenths gave the best classification results. The percentage of correctly classified were 70.8 % and 70.7 % for the calibration and validation sets, respectively. The ANN models with 10 hidden layers and 0.4 learning rate were developed. And those for model developed by ANN were 90.9 % and 89.3 %, respectively. The performance of nolinearANN model was better than the other two models.4,The models for reducing sugar, glucose, fructose, moisture, sucrose, maltose, diastase activity, acidity and hydroxymethyl furfural were established by partial least squares (PLS) and multiple linear regression (MLR) in 11800-4100 cm^-1, respectively. The results demonstrated that the ratio of prediction to deviation (RPD) of the model developed by PLS for the components were 2.44, 1.85, 2.32, 1.52, 1.24, 1.40, 1.87, 1.46 and 1.21, respectively. And those for model developed by MLR were 2.26, 1.86, 2.32, 1.50, 1.69, 1.07, 1.27, 1.70 and 1.49, respectively. The prediction accuracy of models for reducing sugar, glucose, fructose and water are better than other indicators.5,The performances of models for determination of reducing sugar, glucose and fructose developed by BiPLS, SiPLS together with optimization conditions were compared. For the BiPLS models of the three indictors, the R were 0.913, 0.903 and 0.881, the RMSEP were 1.92, 1.73 and 1.15 and the RPD were 2.06, 2.01 and 2.28, respectively. Those for the SiPLS models, the R were 0.916, 0.921 and 0.902, the RMSEP were 1.77, 1.73 and 1.16 and the RPD were 2.24, 2.01 and 2.26, respectively. The models for those indicators were also developed by ANN ,with the R of 0.916, 0.926 and 0.921, the RMSEP of 1.53, 1.37 and 1.13 ,and the RPD of 2.59, 2.53 and 2.32, respectively. Compared the resluts of the two models, the prediction performance of non-linear models was better than linear models.6,The use of fiber optic transreflectance near infrared spectroscopy (NIR) in combination with DM, MD-DA and DPLS chemometric techniques has been investigated to discriminate the authenticity of honey. The identification models were constructed to classify the pure honey and the adulterated honey samples with C4 plant sugar by DM, MD-DA and DPLS, respectively. The results showed that the discrimination model developed by DM together with SNV+1D was best. The total correct classifications of the model was 87.1 % and the correct identification rate of pure honey and adulterated were 75 % and 98.6 %, respectively. The discrimination model of MD-DA together with raw spectra gave the correct classificatoin of 90.0 %, with 100 % and 81.1 % of samples correctly classified for pure and adulterated honey, respectively. DPLS together with first derivative,mean center and 13 smoothing gave the correct classification of 89.1 %. Adulteration honey samples were correctly classified (97.7 %) and pure honey achieve a correct classification of 82 %.