Study On Data Processing Of Raman Spectrum Based On Mini-spectroscopy

Raman spectrum is a kind of molecule scattering spectroscopy, which is characterized by the frequency excursion that caused by interactions of molecule and photon to show the information of molecule. Every substance has its own feature Raman spectrum, which intensity is proportional to its concentration under certain condition. This is the basis of Raman spectroscopy to analyze the structure, components, concentrations and some other properties of samples. There are many advantages to non-destructive analyze with Raman spectrum, such as non-osculatory, non-destructive, high sensitivity, short time ,a spot of sample and non-preparative sample. So Raman spectroscopy has became one of research hotspot of analytical science field. This thesis researched the data disposal of Raman spectroscopy. The main contributions of this thesis are as follows:1. Explain the Raman scatter effect by classical theory and quantum theory, respectively. And introduce detailedly the development and application of Raman spectral analysis technology, and the configuration of Raman spectroscopy.2. Because these some common pre-processing methods can not by a long sight settle fluorescence background interferer, so apply wavelet transform in the pre-processing of Raman spectra in order to reduce the fluorescence backgrounds and to improve the signal-to-noise ratio of the spectra. Experimental results show that the application of wavelet transform obviously improves the prediction performance of Partial Least Squares-Support Vector Machine model of Raman spectrum.3. Because of the lack of traditional detection, wavelet transform multi-scale singularity detection is applied. Experimental result shows that it can determine executively the place of characteristic point of complicated pecks include acromion, and powerful ability of ant-noise.4. For quantitative analysis, a new modeling method Partial Least Squares-support vector machine (PLS-SVM) is introduced to the quantitative calibration of Raman spectroscopy. First of all, because it extracts data character of input spectrum by using PLS, so the dimensions of input variable of SVM modeling are decreased effectively. At the same time the PLS-SVM model preserves favorable performance of SVM model. Experimental results show that the forecast effect of PLS-SVM model precedes obviously the common PLS model.