The Applied Research Of Near-infrared Micro- Spectrometer In The Detection Of Liquor

As an rising analysis technology, near-infrared spectroscopy (NIRS) has many advantages, such as not destroying samples, not using solvent, no need to pretreatment, predicting several components at a time and fast calibration speed, NIRS has been widely used in the fields of food, pharmaceutical, agriculture, petrochemicals, environment and many others. Based on near-infrared micro-spectrometer developed by Micro-system Research Center of Chongqing University, this paper presents its application in component analysis and classification of various liquors.Based on the near-infrared spectral data, the qualitative and quantitative models of liquor are constructed using chemometrics methods, such as principal component analysis (PCA), partial least squares (PLS), least square support vector machines (LSSVM), and several pattern recognition methods, to perform component analysis and the classification. The work of this paper is summarized as follows:①Firstly, NIR experiment platform around the near-infrared micro-spectrometer is set up. The spectra of several alcohol solutions with different concentrations and six brands of liquors in the market are collected. In order to improve the generalability and stability of the calibration models, several algorithms for pretreatment are evaluated.②The principal component analysis and partial least-squares are used to extract the relevant feature for modeling from the spectra, and then the pattern recognition methods (such as SIMCA, Mahalanobis distance) are applied to construct a linear qualitative model. Taking into account the possible non-linear relationship between the spectra and concentrations, we use the least squares support vector machine to construct a non-linear qualitative model. The results show that the least squares support vector machine has the best classification rate.③The quantitative model of liquors is constructed using the spectra of alcohol solutions. Then the principal component analysis, partial least squares and least squares support vector machine are compared when modeling. The results show that the least squares support vector machine model results in a RMSEC of 0.0040 and RMSEP of 0.2912. This result is superior to that of principal component analysis and partial least squares model.