Research On Determination Of Main Components In Milk By Short-wave Near-infrared Spectroscopy

With the advantages of rapid detection, low cost, nondestructive testing and reagent-free analysis, the near-infrared spectroscopy analysis (NIRS) technique is not only suitable for laboratory analysis, but also for in-situ real-time analysis and process control. In this thesis, the short-wave near-infrared spectroscopy (SWNIRS) technique was applied to the determination of main components (fat, protein and lactose) in milk. With relevant chemometrics methods, several key problems were solved, such as sensitive wavelength selection, spectral data pretreatment, model optimization and instrument development. The main works of this thesis are as follows:The near-infrared transmission spectrum (600-1100nm) of 78 milk samples manufactured by different dairy companies were collected by the common NIR scanning spectrometer, and the concentration of the main components in samples mentioned above were analyzed according to the relevant GB standards. The UVE-PLS method was used to select the sensitive wavelength. The LEDs and narrow-band interference filters were chose and fitted together with a special elliptical shell to make a light-focus light source. The optical path was optimized and a novel high-precision current mirror circuit was developed to ensure the stability of the power supply for the light source. The photoelectric circuit and the control system based on a ARM microprocessor were designed and optimized. The embedded software of the analyzer was developed with the GNU arm-elf-toolchain, and the calculation programs for PC were developed with the GNU octave V3.0. Finally, total 2 instruments were built and tested.89 milk samples collected from several markets of Changchun and Beijing were analyzed with the analyzer mentioned above. The RHM method was applied to find and remove the outliers before the PLS model was built. The leave-one-out method was applied to get the principal factor number during model building. The root mean square error of prediction (SEP) values of fat, protein lactose were 0.1074,0.1837 and 0.3273, and the coefficient of correlation (R) were 0.970, 0.961, and 0.872.