Application of Raman Spectroscopy and Pattern Recognition Methods for Determining the Authenticity and Detecting the Adulteration of Milk Powder
WANG Hai-yan1, 2, SONG Chao1, LIU Jun1, 2, ZHANG Zheng-yong1, 2, XIE Wei-liang1, 2, LI Li-ping3, SHA Min1,2*
1. School of Management Science & Engineering, Nanjing University of Finance & Economics, Nanjing 210046, China 2. Jiangsu Province Institute of Quality & Safety Engineering, Nanjing 210046, China 3. School of Mechanical Engineering, Nanjing University of Science & Technology, Nanjing 210046, China
摘要： 奶粉的真伪和掺伪近年来受到广泛的关注，研究一种操作便捷，能准确、快速、全面鉴定奶粉品牌并实现奶粉掺假鉴别的新方法对于奶粉的质量控制具有重要的意义。为实现奶粉的真伪鉴别，采集三种品牌奶粉贝因美、飞鹤和雀巢的拉曼光谱，并利用拉曼谱图特征峰结合最近邻算法(nearest neighbor，NN)的模型对三种品牌奶粉进行识别，在10次交叉验证的基础上，平均识别率为99.56%。为实现奶粉的掺伪分析，将飞鹤奶粉与雀巢奶粉按不同质量比(0∶1，1∶3，1∶1，3∶1，1∶0)混合成五种掺伪奶粉，提取掺伪奶粉中的脂肪，采集脂肪样本的拉曼光谱，分别使用拉曼谱图特征峰结最近邻算法的模型和核主成分分析(kernel principal components analysis，KPCA)结合最近邻算法的模型对五种脂肪样本进行识别，10次交叉验证下的平均识别率分别为93.33%和98.89%，平均运算时间分别为0.085和0.104 s。实验证明：特征峰结合NN的算法可以快速实现对奶粉真伪的判别，但此算法不能很好的区分掺伪奶粉;拉曼光谱-KPCA-NN模型可以为奶粉的掺伪检测提供一种简便、准确、快速的方法。
关键词：奶粉;拉曼光谱;核主成分分析;最近邻算法;真伪;掺伪
Abstract：The authenticity and adulteration of dairy products are attracting broad attention in recent years. There is a need to develop rapid, simple and accurate analytical methods for the detection of authenticity and adulteration of dairy products. To discriminate between milk powder samples, Raman spectra of FIRMUS, Nestlé and Being Mate milk powder were collected. The nearest neighbor algorithm (NN)combined with the characteristic peaks were employed for the design of a model. On the basis of 10 cross validation, the average recognition rate was 99.56%. In order to achieve the analysis of the adulteration of milk powder, FIRMUS milk powder was mixed with Nestlé milk powder according to the mass ratio 0∶1, 1∶3, 1∶1, 3∶1 and 1∶0 to get five kinds of the adulterated milk powder samples. Then, fat was extracted from the adulterated milk powder samples. Raman spectra of the fat were collected, then two methods were employed for the design of models. One was the nearest neighbor algorithm combined with the characteristic peaks, another was the kernel principal component analysis (KPCA) combined with NN. On the basis of 10 cross validation, the average recognition rate reached 93.33% and 98.89%, the average operation time was 0.085 and 0.104 s. The results of this work showed that the nearest neighbor algorithm combined with the characteristic peaks can be applied for the determination of the authenticity of milk powder while Raman-KPCA-NN model can provide a simple, accurate and rapid method to investigate the adulteration of milk power.
[1] Mohammad K, Romdhane K. Trends Food Sci. Technol., 2015, 46(1): 27. [2] Santos P M, Pereira-Filho E R, Rodriguez-Saona L E. Food Chem., 2013, 138(1): 19. [3] Hrbek V, Vaclavik L, Elich O, et al. Food Control, 2014, 36(1): 138. [4] McGoverina C M, Clarkb A S S, Holroydc S E, et al. Anal. Chim. Acta, 2010, 673(1): 26. [5] LI Bing-ning, WU Yan-wen, WANG Yu, et al(李冰宁, 武彦文, 汪 雨，等). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2014, 34(10): 2696. [6] Turk M, Pentland A. Journal of Cognitive Neuroscience, 1991, 3(1): 71. [7] HAN Ping, WANG Huan(韩 萍, 王 欢). Journal of Processing(信号处理), 2013, 29(12): 1697. [8] Sepehr D, Mobarakeh A K, Suandi S A, et al. Procedia Engineering, 2012, 41: 516. [9] Ión F A, Garrigues S, Guardia M. Anal. Chim. Acta, 2004, 513(2): 401. [10] Wu D, He Y, Feng S J, et al. J. Food Eng., 2008, 84(1): 124. [11] Kasemsumran S, Thanapare W, Kratsoonthan A. Anal. Sci., 2007, 23(7): 907. [12] Almeida M R, Oliveira K S, Stephani R, et al. J. Raman Spectrosc., 2011, 42(7): 1548. [13] Li-Chan E C Y. Trends Food Sci. Technol., 1996, 7(11): 361. [14] Zhou Q, Sun S Q, Yu L, et al. J. Mol. Struct., 2006, 799: 77. [15] Benzerdjeb A M, MokhtariI N T, Rahal M S. Spectrochim. Acta, Part A, 2007, 68(2): 284.
