胭脂红是一种应用广泛的食品色素, 在各种食品、 饮料的添加剂里都有它的身影, 过量食用人工合成色素会严重危害健康。 食物中色素一般都是多种联用, 各种色素之间会相互产生干扰, 这加大了对食品中色素检测的难度, 模拟食品中多种色素共存的环境, 采用荧光光谱技术, 结合PSO-SVM算法, 建立一种测定混合溶液中胭脂红含量的方法。 从试剂公司购买胭脂红和苋菜红固体粉末, 选择胭脂红为待检测色素, 苋菜红为干扰色素, 配成不同浓度的胭脂红单色溶液以及加入苋菜红后的混合溶液样本, 其中胭脂红的浓度在0.1~30 μg·mL-1之间, 干扰色素苋菜红的浓度在0.1~10 μg·mL-1之间随意添加。 运用Edinburgh Instruments 公司生产的FS920稳态荧光光谱仪, 测得胭脂红单色溶液与加入苋菜红后混合溶液的荧光光谱图, 分析得到胭脂红的最佳激发波长为λex＝326 nm, 最佳发射波长为λem＝430 nm。 各选取6组不同浓度的单色样本以及混合色素样本, 其中, 胭脂红的物质浓度同为3, 4, 5, 6, 7和8 μg·mL-1, 苋菜红的物质浓度都定在2 μg·mL-1。 观察6组样本在激发波长λex=326 nm时的发射光谱和荧光强度的关系。 单色样本中, 胭脂红浓度与荧光强度线性关系良好; 而在混合溶液中, 随着胭脂红浓度的增加, 荧光强度呈现出先降后增再降的过程, 光谱线型、 强度与各组分浓度间存在复杂的非线性关系, 得以证明混合溶液的荧光光谱并不是由各组分光谱简单的叠加, 而是在吸收光谱的过程中, 胭脂红溶液与苋菜红溶液存在竞争和相互影响。 配取25组胭脂红、 苋菜红混合溶液, 从中选择7个作为预测样本, 其余18组作为训练样本。 7 个预测样本中胭脂红的浓度分别为 1.0, 2.0, 4.0, 6.0, 9.0, 12和15 μg·mL-1, 干扰物质苋菜红的物质浓度在0.1~10 μg·mL-1之间。 选择各组样本在最佳激发波长λex＝326 nm 下对应的荧光强度, 作为检测模型的输入, 以胭脂红的预测浓度作为输出。 对PSO参数初始化设置后, 训练输出SVM的最佳参数c和g, 将所得的最佳参数输入PSO-SVM模型, 得到7组预测样本的浓度预测结果分别为: 1.146 9, 1.860 6, 3.854 4, 6.146 9, 9.133 8, 11.857 6和14.859 8 μg·mL-1。 分析PSO-SVM的预测结果, 得到胭脂红平均回收率为100.84%, 预测均方根误差（RMSEP）为1.03×10-4, 模型输出与真实值之间的相关系数是0.999。 在同等条件下, 采用误差逆向传播算法（BP）预测得到的7组样本浓度分别为: 1.140 1, 2.139 8, 3.188 2, 6.436 2, 8.882 7, 11.860 1和12.664 3 μg·mL-1, 其平均回收率为98.56%, 均方根误差为4.65×10-3, 输出值与真实值之间的相关系数为0.972。 与误差逆向传播算法（BP）的预测结果相比较, PSO-SVM 相关系数高出2.7%, 平均回收率高出0.6%, 均方根误差降低了将近一个数量级。 分析结果表明, 通过荧光光谱技术与PSO-SVM相结合的方法, 能够有效的避开干扰色素的影响, 准确的测定混合溶液中胭脂红的含量, 并且效果相比较于BP更加理想。

Carmine is a widely used food pigment in various food and beverage additives. Excessive consumption of synthetic pigment shall do harm to body seriously. The food is generally associated with a variety of colors. Various pigments will interfere with each other, which increases the difficulty of detection of pigments in food. Under the simulation context of various food pigments’ coexistence, we adopted the technology of fluorescence spectroscopy, together with the PSO-SVM algorithm, so as to establish a method for the determination of carmine content in mixed solution. Carmine and amaranth solid powders were purchased from reagent company. Carmine was selected as pigment to be detected, and amaranth was interfered pigment, carmine monochromatic solution with different concentrations and mixed solution after adding amaranth. The carmine concentrations of 0.1~30 μg·mL-1, interfered pigment amaranth concentrations of 0.1~10 μg·mL-1 were arbitrarily added. Using the FS920 steady state fluorescence spectrometer produced by Edinburgh Instruments Company, the fluorescence spectra of the carmine monochromatic solution and the mixed solution after the addition of amaranth were measured. The optimal excitation wavelength of carmine was λex=326 nm. The optimal emission wavelength For λem=430 nm. The six different concentrations of monochromatic samples and mixed pigment samples were selected. Among them, the concentration of amaranth was set at 2 μg·mL-1, and the concentration of carmine was 3, 4, 5, 6, 7, 8 μg·mL-1. Observe the relationship between the emission spectra and the fluorescence intensity of the six samples at the excitation wavelength λex=326 nm. In the monochromatic samples, the carmine concentration and fluorescence intensity were linear well. The fluorescence intensity of the six samples decreased first and then increased and then decreased again with the increase of the carmine concentration. It is proved that the fluorescence spectrum of the mixed solution is not simply superimposed by the spectrum of the components, but rather the competition and interaction between the carmine solution and amaranth solution in the process of absorbing the light spectrum. With 25 sets of carmine and amaranth mixed solution, seven of them were selected as prediction samples and the remaining 18 groups were used as training samples. The concentrations of carmine in the seven predicted samples were 1.0, 2.0, 4.0, 6.0, 9.0, 12 and 15 μg·mL-1, and the concentrations of the intercalating matter amaranth in the range of 0.1~10 μg·mL-1. The fluorescence intensities corresponding to the optimal excitation wavelength λex=326 nm of each sample were selected as the input of the detection model, and the predicted concentration of carmine was taken as the output. After initializing the PSO parameters, the optimal parameters c and g of SVM were trained. The optimal parameters were input into the PSO-SVM model. The predicted results of the seven predicted samples were: 1.146 9, 1.860 6, 3.854 4, 6.146 9, 9.133 8, 11.857 6, 14.859 8 μg·mL-1. The results of PSO-SVM analysis showed that the average recovery of carmine was 100.84%, and the root mean square error of prediction (RMSEP) was 1.03×10-4, and the correlation coefficient between model output and real value was 0.999. Under the same conditions, the concentrations of seven samples predicted BP method were 1.140 1, 2.139 8, 3.188 2, 6.436 2, 8.882 7, 11.860 1 and 12.664 3 μg·mL-1. The average recoveries was 98.56% The RMSEP was 4.65×10-3 and the correlation between the output and the true value was 0.972. Compared with the predicted results of reverse transmission, the correlation coefficient of PSO-SVM was 2.7% higher, the average recovery rate for 0.6%, and the root mean square error was nearly one order of magnitude lower. According to the analysis results, it can effectively avoid the interference caused by pigment with the combination of the fluorescence spectrum technique and PSO-SVM, accurately determining the content of carmine in mixed solution with an effect better than that of BP.