Research On Spectral Multivariate Calibration Model

Infrared spectroscopy technology which used as a practical techniques has been greatly developed in recent year. There are currently widely used in many fields, such as agriculture, medicine, chemistry, petrochemicals and so on. Infrared spectroscopy technology is an indirect measurement method, which need to build the model for analysis. Depending on the purpose of analysis, the infrared spectroscopy technology can be divided into qualitative analysis and quantitative analysis. But with the promotion of infrared analysis technology, common analysis methods cannot achieve good results in some complex analytical problems. Based on the issues in current infrared spectroscopy applications, this paper will research on both qualitative analysis and quantitative analysis.Infrared spectroscopy qualitative analysis is to establish the relationship between the sample category and spectral. We can use the relationship to predict the category of unknown samples. In this paper, we research the qualitative analysis when the ingredients are very close between different samples. We know that infrared spectrum is a reflection of the molecular structure. So it is difficult for qualitative analysis when the infrared spectrum closed. The dimension often is very high when establish the analysis model, so we should reduce the dimension firstly. The researchers often use principle component analysis (PCA) to reduce dimension currently. But PCA is mainly retained the largest original information, so the classified information is often missed. Therefore, we used the KL to reduce the data dimension. The data after KL can remain more classified information. And then the BP neural network be constructed using the data after KL. Compared with the common qualitative analysis method, the KL-BP model which proposed in this paper has the best classification results in 9 kinds of edible oil.In the quantitative analysis, my research is mainly on establishing the model in small data set. In practice, we often can obtain little sample which restricted the sample source and other factors. And we will spend a lot of time to measure the spectrum when we have a large number of samples. In this paper, I study on the infrared spectroscopy quantitative analysis in small samples situation. Firstly, we expand the number of samples by Bootstrap resampling method which using the original samples. And then the noise injection can be used to improve the diversity of these samples. Finally, the quantitative analysis model can be established by SVM (Support vector machine). Compared with other analysis method, the Bootstrap-SVM model can get the best result by using small samples.