[Application of three-dimensional fluorescence spectra technique to discrimination of distilled spirits based on...

In order to study the identifying and classification method of distilled spirits, about 100 kinds of normative distilled spirits were selected as the laboratory samples and their fluorescence spectra induced by ultraviolet light were measured by Roper-Scientific SP-2558 respectively. Then, using the statistics and plot software Origin 7.5, the authors composed the three-dimensional fluorescence spectra of the distilled spirits. In the meanwhile, the three-dimensional fluorescence spectra of the dis tilled spirits were also studied respectively. The authros compared the three-dimensional fluorescence spectra of different distilled spirits with the same brands and that of different spirits with the same brands. The authors are very confident that there must be some typical parameters, by which the authors can distinguish different kinds of distilled spirits with different brands effectively. And the authors found that while the three-dimensional fluorescence spectra of different distilled spirits with different brands are definitely different by their three typical parameters including the number of the major summit, position of the major summit and optimum wavelength, the three typical parameters of the three-dimensional fluorescence spectra of different distilled spirits with the same brand are very similar. Finally, the authors extracted the three typical parameters that are the number of the major summit, position of the major summit and optimum wavelength. Then, the data were processed by computer for emulation. As a result, the authors found that, using the three characteristic parameters, different kinds of distilled spirits can be classified ac curately by the algorithm of LS-SVM (least-squares support vector machine). The outcomes of the data emulated by different algorithms were obtained seriatim. The authors compared the outcomes, and the fact proved that more accurate outcome of identifying and classification can be obtained by LS-SVM.