|
This thesis describes chemometric methods and their use in near infrared spectroscopy (NIR). The particular applications in this study include NIR analyses of Chinese fir and milk powder using transmittance near infrared spectrophotometers. Using the optimum model for the determinationof of the prediction samples. Different calibration techniques, including back-propagation neural network (BP-ANN), radial basis function networks (RBF-NN), support vector machines (SVM), partial least squares regression (PLS)were investigated and applicability for the prediction of the content of Chinese fir and milk powder was discussed. In addition, the techniques for calibration subset selection and mathematical sample pre-treatments were explored. In this research, chemometrics methods are applied in the prediction of the content of Chinese fir and milk powder. The results as follows:1. The amount of holocellulose, lignin, density, and microfibril angle of Chinese fir were predicted by using back-propagation neural network (BP-ANN) combined with near infrared (NIR) spectrometry. First, the data of original spectra were pretreated by Savitzky-Golay smoothing algorithm and the second derivative, then the data of near infrared spectrometry with 171 point was compressed to 86 by using wavelet transform, finally, the models were established by using BP-ANN. The influences of the number of hidden neurons, learning rate, Momentum, and epochs were discussed in this paper, and the models were validated using leave-n-out cross-validation approach. The prediction samples, which were not used in the model generation, were predicted by using the obtained models, the correlation coefficients (R) of holocellulose, lignin, density, and microfibril angle were 0.91、0.90、0.78, and 0.87, respectively. The standard errors of prediction (SEP) of the established models were 0.86%, 0.33%, 2.29%, and 4.99%, respectively. The obtained results showed that the method is fast and nondestructive technique and it can basically satisfy the requirement of quantitative analysis.2. The amount of density、microfibril angle of Chinese fir were predicted by using radial basis function networks (RBF) combined with near infrared (NIR) spectrometry. The raw spectra were pretreated by the second derivative and smoothing, prediction models were established by using RBF, the models were validated using Leave-one-out cross-validation approach, and the influence of parameters was discussed. The correlation coefficients (R) for prediction models of density, microfibril angle of Chinese fir were 0.81, 0.87, respectively. The root mean square errors of prediction (RMSEP) for the established models were 0.03, 0.04, respectively. The results showed that the method can basically satisfy the requirement of quantitative analysis.3. The amount of density of Chinese fir was predicted by using support vector machine (SVM) combined with near infrared (NIR) spectrometry. Fist, smoothing, second derivative and range-scaling were used for the pretreating methods of the NIR spectra of density of Chinese fir.Than the models was established by using SVM, and the effects of parameters were investigated. Under optimized conditions, the correlation coefficient (R) for prediction models of density was 0.93. The root mean square error of prediction (RMSEP) for the established models was 0.017. Comparing with RBF-NN, the predictive results showed that the method proposed was rapid, non-destructive and credible. It was an effective measurement for determining the density of Chinese fir.4. In the present thesis, near infrared spectroscopy (NIRS) with partial least squares (PLS) was applied to determination of melamine content in milk powder. As the results showed, the model based on derivative spectra was not as good as was established based on the original spectra. The most efficacious wavelength range for the determination of melamine content, and it were 4000.63~ 6803.63 cm-1,6800~9700cm-1, respectively. The root mean square error of the calibration set obtained by cross-validation (RMSECV) of the optimum models for the quantitative analysis of melamine content was 0.000498. The correlation coefficient between the predicted values and actual values was 0.99990. Using the optimum model for the determination of melamine content in prediction set, the root mean square error of prediction set (RMSEP) was 0.000221. The correlation coefficient between the predicted values and actual values was 0.92839. These results indicate that it is feasible to use NIR spectroscopy technique for quantitative analysis of melamine content in milk powder.
|