Identification of Adulterants in Adulterated Milks by Near Infrared Spectroscopy Combined with Non-Linear Pattern Recognition Methods
NI Li-jun1, ZHONG Lin1, ZHANG Xin2, ZHANG Li-guo1*, HUANG Shi-xin2
1. Institute of Chemical and Molecular Engineering, East China University of Science and Technology, Shanghai 200237, China 2. Institute of Veterinary Drug Detection, Shanghai Center of Animal Disease Control and Prevention, Shanghai 201103, China
Abstract：In the present work, two hundred and eighty seven raw milks collected from pastures in Shanghai and surrounding areas of Shanghai were used as true milk samples and divided into three true milk sets. Five hundred and twenty six adulterated milk samples, which contained dextrin (or starch) mixed with melamine (or urea, or ammonium nitrate), were prepared as six different adulterated milk sets. The concentrations of these adulterants in the adulterated milks were designed to be 0.15%~0.45% (starch or dextrin), 700~2 100 mg·kg-1 (ammonium nitrate), 524~1 572 mg·kg-1 (urea), and 365.5~1 096.5 mg·kg-1 (melamine) to guarantee the protein content of adulterated milks detected by Kjeldahl method not lower than 3%. All the near infrared spectra (NIR) of the samples should have a pretreatment of normal variable transformation (SNV) before they were used to build discriminating models. The three true milk sets and six adulterated milk sets were combined in different ways in order to build NIR models for discriminating different kinds of adulterants (i. e., dextrin, starch, melamine, urea and ammonium nitrate) based on simplified K-nearest neighbor classification algorithm (IS-KNN) and an improved and simplified of support vector machine (ν-SVM) method. The relationship between mass concentration of the adulterants and the rate of correct discrimination was also investigated. The results show that the average discrimination accuracy of IS-KNN and ν-SVM for identifying melamine, urea and ammonium nitrate were in the region of 49.55% to 51.01%, 61.78% to 68.79% and 68.25% to 73.51%, respectively. Therefore within the concentration regions designed in this study, it is difficult to distinguish different kinds of pseudo proteins by NIR spectroscopy. However, the average accuracy of IS-KNN and ν-SVM for identifying starch and dextrin are 92.33% and 93.66%, 77.29% and 85.08%, respectively. Most discrimination results of ν-SVM are better than those of IS-KNN. The correlative analysis between the discrimination accuracy rate and the content levels of the adulterants indicated that near infrared spectroscopy combined with non-linear pattern recognition methods can distinguish dextrin and starch in milks with higher concentration levels (>0.15%), but do not work well on identifying the adulterants with lower concentrations such as melamine (365.5 to 1 096.5 mg·kg-1), urea (524 to 1 572 mg·kg-1), ammonium nitrate (700 to 2 100 mg·kg-1). Therefore near Infrared Spectroscopy is not suitable for identifying the adulterants with concentrations are below 0.1%.
Key words：Near-infrared spectroscopy;Discrimination of adulterated substances in liquid milk;Improved and simplified K-nearest neighbor classification algorithm;Improved and simplified of support vector machine
[1] Balabin R M，Smirnov S V. Talanta，2011，85(1)：562. [2] Kasemsumran S, Thanapase W, Kiatsoonthon A. Anal. Sci., 2007, 23(7): 907. [3] XU Yun，WANG Yi-ming，WU Jing-zhu，et al(徐 云，王一鸣，吴静珠，等). Journal of Infrared and Millimeter Waves(红外与毫米波学报)，2010，29(1)：53. [4] Yang R, Liu R, Kexin X. Food Bioscience, 2013, 2: 61. [5] Santos P M，Pereira-Filho E R，Rodriguez-Saona L E. Journal of Agricultural and Food Chemistry，2013，61(6)：1205. [6] Santos P M，Pereira-Filho E R，Rodriguez-Saona L E. Food Chemistry，2013，138(1)：19. [7] ZHONG Zhen-zhen，ZHANG Li-guo，ZHANG Xin，et al(钟珍珍，张立国，张 鑫，等). Computers and Applied Chemistry(计算机与应用化学)，2010，27(12)：1691. [8] Ni L J，Zhang L G，Xie J，et al. Analytica Chimica Acta，2009，633(1): 43. [9] Zhang L G，Zhang X，Ni L J，et al. Food Chemistry，2014，145：342. [10] Chan E Y Y，Griffiths S M，Chan C W. Lancet，2008，372(9648)：1444. [11] Schlkopf B，Smola A J，Williamson R C，et al. Neural Computation，2000，12(5)：1207.
