-
为便于神经网络的应用,需对荧光光谱数据进行归一化处理。利用下式将数据归一化到0.1~0.9:
$ y = 0.1 + 0.8\left( {\frac{{x - {x_{{\rm{min}}}}}}{{{x_{{\rm{max}}}} - {x_{{\rm{min}}}}}}} \right) $
(1) 式中, x, xmax, xmin分别表示光谱数据、光谱数据中的最大值和最小值; y表示归一化后的数据[20]。
归一化后的数据作为RBF神经网络的输入。RBF神经网络的输出为胭脂红的浓度。实验中采用交叉验证方法中的留一交互验证法,即对于n个样本,每次剔除一个作为预测样本,其余作为训练样本。实验共预测12次,11组作为训练样本,1组预测,12组样本均被预测一次。采用留一交互验验证法,可充分利用已有的样本,减少神经网络因训练样本的输入顺序不同而产生的随机性,减小误差,提高网络的稳定性[20]。以12组样本的平均相对误差,即相对误差绝对值的平均,来判定模型的准确性。与普通3维荧光光谱数据结合RBF神经网络算法预测结果对比,验证3维同步荧光光谱结合RBF神经网络,测定混合色素溶液中胭脂红的浓度具有优越性。
-
5μg/mL胭脂红、日落黄单体色素溶液,在最佳激发波长下的发射谱如图 1所示。
Figure 1. Emission spectra of carmine and sunset yellow in 5μg/mL solution at the best excitation wavelength
实验中得到胭脂红、日落黄的荧光光谱特征参量如表 1所示。
Table 1. Features of fluorescence spectra
pigment effective excitation wavelength range/nm optimum excitation wavelength/
nmfluorescence emission wavelength range/
nmfluorescence peak wavelength/
nmcarmine 260~350 336 400~510 450 sunset yellow 264~308 288 380~470 418 由胭脂红、日落黄在其最佳激发波长下的发射谱及其荧光特征参量表均可以看出,胭脂红与日落黄荧光峰的位置比较接近。
图 2为5μg/mL胭脂红溶液3维同步荧光光谱等高线图及普通3维荧光光谱等高线图。
图 2a和图 2b中荧光峰的个数相同,荧光特征参量相近。3维同步荧光光谱等高线图中光谱明显发生了窄化。
2维同步荧光光谱和普通3维荧光光谱分别结合RBF神经网络,计算得到的混合溶液中待测物胭脂红的预测浓度和相对误差分别见表 2。表中第1组、第2组样本胭脂红浓度均为2.04μg/mL,干扰物日落黄浓度不等。因表格空间限制,synchronous fluorescence代表3维同步荧光光谱。
Table 2. Relative error and predicted concentration of carmine by 3-D synchronous fluorescence using RBF
actual concentration/
(μg·mL-1)predicted concentration of synchronous fluorescence/
(μg·mL-1)predicted concentration of three dimensional fluorescence/
(μg·mL-1)relative error of synchronous fluorescence prediction/
%relative error of three dimensional fluorescence prediction/
%2.04 2.014 2.150 -1.28 5.37 2.04 2.049 1.655 0.45 18.88 2.72 2.613 2.687 -3.93 -1.22 3.20 3.069 3.294 -4.11 2.95 3.52 3.408 3.133 -3.18 -11.00 3.76 3.776 3.896 0.43 3.61 3.82 3.822 3.793 0.05 0.70 4.3 4.362 3.453 1.44 -19.70 4.36 4.270 4.674 -2.08 7.19 4.52 4.197 5.434 -7.16 20.22 4.80 4.832 4.534 0.67 -5.55 5.26 4.760 3.315 -9.51 -36.98 average relative error/% 2.86 11.12 由表 2可知,3维同步荧光光谱结合RBF神经网络,测定胭脂红浓度,预测的平均相对误差为2.86%。普通3维荧光光谱结合RBF神经网络,预测结果平均相对误差为11.12%,3维同步荧光光谱平均相对误差减小了8.26%。图 3中示出了3维同步荧光光谱、普通3维荧光光谱结合RBF神经网络预测胭脂红浓度结果。x轴为实际浓度,y轴为预测浓度。从图 3可以看出, 3维同步荧光光谱预测结果均匀分布在直线y=x两侧,距离较近,表明预测结果较好。普通3维荧光光谱预测结果距离直线分布较远,表明预测结果较差。
-
(1) 同步荧光分析法能够减小光谱重叠、窄化谱带、提高分辨率,对于多组分混合物的分析常采用该分析方法。3维同步荧光光谱中Δλ为一个范围,与2维同步荧光光谱相比,有效避免了有用荧光信息的丢失。可知,对混合色素溶液中各成分含量的检测,3维同步荧光光谱更具有优势。
(2) 3维同步荧光光谱数据为非线性结构,RBF神经网络适用于非线性模型。因此,测定日落黄、胭脂红两种荧光峰相近的混合色素溶液中胭脂红的浓度,3维同步荧光光谱结合RBF神经网络结果较好。
应用3维同步荧光光谱测定胭脂红浓度
Determination of carmine concentration with 3-D synchronous fluorescence spectrometry
-
摘要: 为了测定混合色素溶液中胭脂红的浓度,采用归一化的方法对荧光光谱进行数据预处理,将处理后的光谱数据结合径向基神经网络,建立了对胭脂红含量的预测模型。结果表明,3维同步荧光光谱、普通3维荧光光谱预测结果的平均相对误差分别为2.86%,11.12%;对于混合色素溶液中单个色素浓度的测定,3维同步荧光光谱结合径向基神经网络效果较好。该研究为预测混合色素溶液中各色素浓度提供了帮助。Abstract: In order to determine carmine concentration in the mixed pigment solution, normalization method was used to preprocess the fluorescence spectra. The processed data were combined with radial basis function neural network to establish the prediction model of carmine content. The average relative error of the prediction results of 3-D synchronous fluorescence spectrometry and 3-D ordinary fluorescence spectrometry were 2.86% and 11.12% respectively. The results showed that the 3-D synchronous fluorescence spectrometry was superior for the determination of the mixed pigment solution. The research provides the help for the prediction of the pigment concentration in the mixed pigment solution.
-
Table 1. Features of fluorescence spectra
pigment effective excitation wavelength range/nm optimum excitation wavelength/
nmfluorescence emission wavelength range/
nmfluorescence peak wavelength/
nmcarmine 260~350 336 400~510 450 sunset yellow 264~308 288 380~470 418 Table 2. Relative error and predicted concentration of carmine by 3-D synchronous fluorescence using RBF
actual concentration/
(μg·mL-1)predicted concentration of synchronous fluorescence/
(μg·mL-1)predicted concentration of three dimensional fluorescence/
(μg·mL-1)relative error of synchronous fluorescence prediction/
%relative error of three dimensional fluorescence prediction/
%2.04 2.014 2.150 -1.28 5.37 2.04 2.049 1.655 0.45 18.88 2.72 2.613 2.687 -3.93 -1.22 3.20 3.069 3.294 -4.11 2.95 3.52 3.408 3.133 -3.18 -11.00 3.76 3.776 3.896 0.43 3.61 3.82 3.822 3.793 0.05 0.70 4.3 4.362 3.453 1.44 -19.70 4.36 4.270 4.674 -2.08 7.19 4.52 4.197 5.434 -7.16 20.22 4.80 4.832 4.534 0.67 -5.55 5.26 4.760 3.315 -9.51 -36.98 average relative error/% 2.86 11.12 -
[1] NATIONAL HEALTH AND FAMILY PLANNING COMMISSION OF THE PEOPLE'S REPUBLIC OF CHINA.GB2760-2011 National food safety standards-standards for uses of food additives [S].Beijing: Standards Press of China, 2011: 8-179(in Chinese). [2] CHEN G Q. Studies on application of fluorescence spectroscopy in food safety supervision[D]. Wuxi: Jiangnan University, 2010: 1-2 (in Chinese). [3] VAVROUS A, VAPENKA L, SOSNOVCOVA J, et al. Method for analysis of 68 organic contaminants in food contact paper using gas and liquid chromatography coupled with tandem mass spectrometry[J]. Food Control, 2016, 60:221-229. doi: 10.1016/j.foodcont.2015.07.043 [4] XIA L Y, HAN Y Y, KUANG L H, et al. Simultaneous determination of basic orange 2, metanil yellow, tartrazine, sunset yellow in bean-product and acid orange Ⅱ, ponceau 2R, rhodanmine B in chili powder by thin-layer chromatographic scanning [J]. Chinese Journal of Analysis Laboratory, 2010, 29(6):15-18(in Chinese). [5] TANG T X, XU X J, WANG D M, et al. A rapid and green limit test method for five synthetic colorants in foods using polyamide thin-layer chromatography[J]. Food Analytical Methods, 2015, 8(2):459-466. [6] VLAJKOVIC J, ANDRIC F, RISTIVOJEVIC P, et al. Development and validation of a tlc method for the analysis of synthetic food-stuff dyes[J]. Journal of Liquid Chromatography & Related Technologies, 2013, 36(17):2476-2488. [7] BLAZHEYEVSKIY N Y, LABUZOVA Y Y. Determination of cefazolin by oscillographic polarography as its S, S'-dioxide[J]. Journal of Analytical Chemistry, 2014, 69(9):883-886. doi: 10.1134/S106193481407003X [8] JIANG Zh L, LI F, LIANG H. Resonance scattering spectroscopic study of the system of PMo12 heteropoly acid and rhodamin S[J]. Acta Chimica Sinica, 2000, 58(8):1059-1062(in Chinese). [9] MEI F, ZHAO X Y, QU F. Comparison of interaction between cytochrome c and single strain deoxyribonucleic acid pools based on capillary electrophoresis[J]. Chinese Journal of Chromatography, 2012, 30(12):1229-1234(in Chinese). [10] REFINETTI P, MORGENTHALER S, EKSTROM P O. Cycling temperature capillary electrophoresis: Aquantitative, fast and inexpensive method to detect mutations in mixed populations of human mitochondrial DNA[J]. Mitochondrion, 2016, 29:65-74. doi: 10.1016/j.mito.2016.04.006 [11] ZHENG C Y, GUO Zh H, JIN L. Measurement of total viable count on chilled mutton surface based on hyperspectral imaging technique[J]. Laser Technology, 2015, 39(2):284-288(in Chinese). [12] WANG X J, YANG L, LIU D L, et al. Determination of vanillic acid in rhizoma picrorhizae and body fluids by excitation-emission matrix fluorescence coupled with second-order calibration methods[J].Journal of Heibei Normal University(Natural Science Edition), 2016, 40(5):412-417(in Chinese). [13] MAO L X, ZHANG X L, FAN S H. Determination of content of raw juice in orange juice by synchronous fluorescence spectrometry[J]. Science and Technology of Food Industry, 2016, 37(6):90-93(in Chinese). [14] WANG Sh T, CHEN D Y, WANG X L, et al. Detection of polycyclic aromatic hydrocarbons combining fluorescence analysis with ABC-BP neural network [J]. Chinese Journal of Lasers, 2015, 42(11):1115001(in Chinese). doi: 10.3788/CJL [15] XU J G, WANG Z B. Fluorescence spectroscopy [M]. 3rd ed. Beijing: Science Press, 2006: 131-160(in Chinese). [16] KUMAR K, MISHRA A K. Parallel factor (PARAFAC) analysis on total synchronous fluorescence spectroscopy (TSFS) data sets in excitation-emission matrix fluorescence (EEMF) layout: Certain practical aspects[J]. Chemometrics and Intelligent Laboratory Systems, 2015, 147:121-130. doi: 10.1016/j.chemolab.2015.08.008 [17] LIU B, GUO H X. MATLAB neural network super learning handbook[M]. Beijing: Beijing University of Posts and Telecommunications Press, 2014: 194-216(in Chinese). [18] SHI F, WANG X C, YU L, et al. 30 case analysis of MATLAB neural network[M]. Beijing: Beijing University of Aeronautics and Astronautics Press, 2011: 65-73(in Chinese). [19] LI R, CHEN G Q, ZHU C, et al. Quantitative analysis of two food colours using excitation-emission matrix spectra coupled with parallel factor algorithm[J]. Spectroscopy and Spectral Analysis, 2014, 34(1):111-115(in Chinese). [20] KONG F B. The study of the synchronous fluorescence of several food colors[D].Wuxi: Jiangnan University, 2013: 31-43(in Chinese).