Study On Several Key Techniques Of Near-Infrared Spectroscopy Analysis

Near-infrared spectroscopy analysis technique is efficient, rapid, noninvasive, environmental friendly and can be run at low costs. It is not only suitable for laboratory analysis, but also in-field fast measurement and real-time on-line analysis. However, there are some key techniques have not been solved thoroughly, which deter further application and development of near infrared spectroscopy (NIRS) technique. In the dissertation, an in-depth study is carried out by the author by synthesizing multi-disciplinary knowledge on the following topics: the abstraction of weak spectral signal, the optimal pathlength measurement condition, the optimization of regression method, the enhancement of multivariate calibration model robustness and the physical explanation of measurement results are studied.The main research content of the dissertation involves:1. The selection principle of optimal pathlength in multivariate calibration is proposed and established. "The method of multiple optimal pathlength combination" to improve prediction accuracy in multivariable calibration is theoretically verified and experimentally validated for the first time. The influence of pathlength selection on prediction accuracy of univariate calibration and multivariate calibration is investigated. The simple determination method of the optimal pathength of water solution is given. 2. The propagation function between the prediction accuracy, instrumental precision, and regression method is established, which has been verified experimentally. The determination method of essential instrumental precision to realize the anticipated prediction accuracy is given. The influence of regression method and the complexity of the sample components on prediction accuracy and essential instrumental precision are discussed. The efficient means to enhance the prediction accuracy is presented.3. The variation of the number of the principle components, shape, physical meaning and the explainablity with the change of calibration data and the sample complexity is systematically studied for the first time. It is pointed out that the principle components are influenced by many factors. The analysis of meaning for the principle components in multivariate calibration should combine with the professional knowledge about measuring sample. The analysis method is not completely fixed and specific for the different measurement object. We can know if the calibration mode is disturbed by the environmental noise by the principle components analysis, which provide the physical basis for the explanation of measurement result and the optimization of the measurement method and the measurement condition.The application of wavelength selection with genetic algorithms is systematically studied. The routine wavelength selection method based on genetic algorithms is modified, which enhance prediction accuracy and improve the robustness of the calibration model. These are all verified by experiments. The wavelength selection approach by the net analyte signal relative error is studied thoroughly. The4. calculation method of net partial sensitivity of unknown sample spectrum in inverse multivariate calibration is put forward, which enable the wavelength selection in practical near-infrared spectroscopy analysis application.5. The calibration model standardization is systematically studied to enhance the robustness and adaptability. Several mathematical pretreatment methods and model transfer methods are analyzed and compared, and some have been verified by experiments. Meanwhile, the effect of the selection of calibration transfer method and the influence of transfer set samples selection on the calibration model transfer result are also discussed.