[Rapid determination of the components in ternary blended edible oil using near infrared transmission spectroscopy].

The FT-NIR transmission spectra of ternary blended edible oil samples were collected over 10 000-4 200 cm(-1). After being pretreated with different methods, the calibration models of quantitative analysis of soybean oil, peanut oil and corn oil contents in ternary blended edible oil were established using partial least square (PLS) regression. The accuracy and precision of the models for the predicted sample set were examined to make sure of the practicability of the models. After being pretreated with first derivative and multiplicative signal correction (FD+MSC), the optimal soybean oil NIR model was built over 5 450.1-4 597.7 cm(-1). The best prediction model for peanut oil was established between 7 521.3 and 6 098.1 cm(-1) after using first derivative with straight line subtraction (FD+SLS) preprocess method. The best pretreated method and the best spectrum range for corn oil content model were first derivative (FD) and 9 993.7-7 498.2 cm(-1), respectively. The best correlation coefficients (R2) of the three prediction models were 99.89%, 99.88% and 99.76%, respectively. The RMSEP of the soybean oil content model was 1.09%, while the peanut oil prediction model's RMSEP was 1.17%, and 1.48% for the corn oil prediction model. The values of the t-test were between 0.007 9 and 0.371 9, and all values of the relative standard deviation (RSD) were less than 1.50%. The results showed that NIR could be an ideal tool for fast determination of the soybean oil, peanut oil and corn oil contents in ternary blended edible oil.

[Application of PCA-SVR to NIR prediction model for tobacco chemical composition].

Near infrared diffuse reflectance spectra of 50 tobacco samples were pretreated with PCA. The calibration models of determination of the main components in tobacco were developed with support v ector regression (SVR). The models weretested with leave-one-out (LOOCV) method and optimized with parameters of kernel function, penalty coefficient C and insensitive loss function. The root mean square errors (RMSE) with leave-one-out cross validation of the optimal models of nicotine, and total sugars, reductive sugar, and total nitrogen were 0.313, 1.581, 1.412 and 0.117 respectively. Based on the comparison of RMSE of the SVM model with those of the partial least square (PLS), multiplicative linear regression (MLR) and back propagation artificial neuron network (BP-ANN) models, it was found that the SVR model was the most robust one. This study suggested that it is feasible to rapidly determine the main components concentrations by near infrared spectroscopy method based on SVR.

[Near infrared determination of the content of caffeine in tea polyphenol].

In the present paper the caffeine in the tea polyphenol was analysed spectrally and quantitatively by using near infrared spectroscopy. From the original absorbance of caffeine in the tea polyphenol an obvious and strong peak can be viewed. By using second derivative, MSC (multiple scatter correction) and correlation analysis the spectral characteristics of caffeine in the near infrared region can be seen very clearly, thus the robust calibration model can be set up easily. The result obtained shows that through this technique the absorptive characteristic of those primary fundamentals of caffeine can be looked through easily, meanwhile, calibration test was performed to quantitatively measure the weight percent of caffeine in the tea polyphenol, and fine precision of the result was obtained in a comparatively very large range of concentration. The SEC(standard error of calibration) is 0.49%, and the correlation coefficient r is 0.993. The result shows that NIR is feasible and superior in analyzing the content of caffeine in tea polyphenol.

[Study on temperature correctional models of quantitative analysis with near infrared spectroscopy].

Effect of enviroment temperature on near infrared spectroscopic quantitative analysis was studied. The temperature correction model was calibrated with 45 wheat samples at different environment temperaturs and with the temperature as an external variable. The constant temperature model was calibated with 45 wheat samples at the same temperature. The predicted results of two models for the protein contents of wheat samples at different temperatures were compared. The results showed that the mean standard error of prediction (SEP) of the temperature correction model was 0.333, but the SEP of constant temperature (22 degrees C) model increased as the temperature difference enlarged, and the SEP is up to 0.602 when using this model at 4 degrees C. It was suggested that the temperature correctional model improves the analysis precision.