Abstract in English:

Several major reference values of wheat flour were modeled using the near-infrared (NIR) spectra of flour samples and the multivariate partial least squares (PLS) method. The calibration models for crude protein, moisture and wet gluten of flour were developed, and the root mean square errors of prediction (RMSEPs) were 0.3784% (w/w), 0.2624% (w/w) and 1.4653% (w/w), respectively. The correlation coefficients squared for prediction (Rp2) were 0.7922 for the best crude protein model, 0.6096 for the best moisture model, and 0.8346 for the best wet gluten model. The results showed that the analysis of the main parameters of wheat flour by benchtop and handheld NIR spectroscopy is feasible and good prediction models can be obtained. Furthermore, the differences in origin of the investigated flour samples (German flour and Chinese flour) can be discriminated by the analysis of their NIR spectra. This discrimination is based on the fact, that the NIR region contains overtone and combination absorption bands of CH, NH, and C=O functionalities which reveal the chemical differences of wheat flour samples of different geographical origin. By using principal component analysis (PCA) German and Chinese flour samples could be well classified by their NIR spectra. However, by using the partial least squares discriminant analysis (PLS-DA) method, the assignment of flour samples to a specific geographical origin can be significantly and effectively improved. In this PhD project, five NIR spectrometers (two benchtop and three handheld systems) are used to measure the NIR spectra of a total of 50 German and 163 Chinese flour samples. The signal-to-noise ratio, spectral resolution, and accuracy of absorbance values of handheld spectrometers are generally lower than those of benchtop spectrometers. According to the near-infrared spectral characteristics of the samples, a PCA model was established, in which IAS 3100 and MicroNIR (VIAVI) achieved 100% correct classification of German and Chinese flour. Furthermore, PLS calibration models for protein, moisture and wet gluten were also established for samples from both countries. Experimental results demonstrate that accurate calibration can be achieved using a benchtop spectrometer with better instrument performance and a handheld spectrometer with more flexible measurement operations. However, calibration models developed using spectra measured with a benchtop spectrometer outperformed those acquired with a handheld spectrometer. In order to demonstrate the transfer of calibration models based on the NIR spectra of different instruments, model transfer methods are discussed in this thesis. For this purpose one spectrometer was defined as master and the others were defined as target instruments and the effects of three spectral standardization methods, direct standardization (DS), piecewise direct standardization (PDS) and simple linear regression direct standardization (SLRDS) algorithms, regarding the sharing of calibration models across instruments were investigated. Applying the three methods, the variability of the spectral data between instruments was significantly reduced and the prediction accuracy of the calibration models for wheat flour parameters was improved.