The Study Of The Synchronous Fluorescence Of Several Food Colors

The synchronous fluorescence spectra of Amaranth, Acid Red and New Red werestudied. The quantitative determination of Amaranth, Acid Red and New Red wascarried out with synchronous fluorescence spectra coupled with curve fitting andartificial neural networks.The three-dimensional fluorescence spectra and the three-dimensional synchronousfluorescence spectra of Amaranth, Acid Red, New Red and the mixture of Acid Redand New Red were measured. The fluorescence spectra was compared with thesynchronous fluorescence spectra and the synchronous fluorescence spectra of themixture were compared with that of Acid Red and New Red, respectively. Therelationship between the intensities of the synchronous fluorescence and theconcentration is studied. The three-dimensional synchronous fluorescence spectra canbe used as an technique of qualitative determination just as the three-dimensionalfluorescence spectra; The fluorescence spectra can avoid the Rayleigh scatteringeffectively; The relation between the intensities and the concentration is nonlinear.The linear curve fitting technique was applied to fit the relationship of thesynchronous fluorescence spectra intensities of Amaranth with its concentration in therange of1ug.ml-1~4ug.ml-1. Three linear fitting models were established based on thesynchronous fluorescence intensities of330nm and542nm. Through comparison, theresults of the mult-variable linear fitting were the best. The mean relative error and themean square error was0.0266and0.0085, respectively. The nonlinear fittingtechnique was used to fit the relationship of the synchronous fluorescence spectraintensities of Amaranth with its concentration in the range of0ug.ml-1~5ug.ml-1.Three models were built. The best one was the mult-variable model. The meanrelative error of it was0.0197and the mean square error was0.0144.Radial basis function neural networks combined with synchronous fluorescencespectra was used to the determination of the Amaranth and the simultaneousdetermination of Acid Red and New Red. The wavelet was applied to deduct the noiseof the synchronous fluorescence spectra. The whole spectra were the inputs of thenetworks. Firstly, the determination of the Amaranth was realized. With the width ofthe Gauss functions was2.95, the mean relative error was0.0123and the the meansquare error was0.0117. Secondly, the simultaneous determination of the Acid Redand the New Red was realized. The networks were built though three steps. Enoughtesting samples were used to fully test the networks. The mean relative error and themean square error was0.0338and0.4745, respectively, when the width of the Gaussfunctions was2.29.The function relationship between the intensities of the synchronous fluorescencespectra of the mixture of the Acid Red and the New Red and their concentrations inthe mixture was discussed through the establishment of the networks. A simple andeffective method for picking up training samples for the radial basis function neuralnetworks was proposed. In the method, the concentration domain of the Acid Red and New Red was divided into nine small areas and a testing sample was placed in eachsmall area. The training samples in each area were decided by the accuracy of itstesting sample. The method was completed in three steps.