基于二维相关谱掺伪牛奶检测方法研究

In view of the serious situation of adulteration in milk, it is urgent to develop a rapid, widely available and cost-effective method to detect the trace adulterant in the complex system of milk for the State Food and Drug Administration, the dairy industry and the consumers. Two-dimensional (2D) correlation spectroscopy is an effective method for three major problems encountered by the conventional spectrum: low selectivity of the spectra, difficulty in extracting the information of the spectral feature and difficulty in spectrogram analysis. In this project, the two-dimensional spectra (IR-IR, NIR-NIR, IR-NIR) are generated from the concentration-perturbed and temperature-perturbed spectral variations of milk adulterated with melamine and urea. The direct pretreatment methods of smoothing and standardization of 2D correlation spectra are developed based on simulated and experimental data. The two mathematical algorithms reconstructing 2D correlation spectra, are studied to mine further meaningful spectral information. The fingerprint databases of pure milk and adulterated milk are constructed，which can potentially be used to discriminate pure milk between adulterated milk comparing their own characteristic peaks. At the same time, the characteristic parameters of the 2D correlation matrix are extracted to establish the eigenvector database of pure milk and adulterated milk, which are combined with pattern recognition for classification of adulterated milk. At the same time, the characteristic parameters of the 2D correlation matrix are extracted to establish the eigenvector database of pure milk and adulterated milk, which are combined with pattern recognition for classification of adulterated milk. Based on 2D and 3D partial least squares methods, the wavenumber range among which spectra information changes obviously with external perturbation, will be selected to construct the models for quantitative analysis of adulterated milk. The influence factors of the models are analyzed. Furthermore, the models are optimized. And this method can be also applied to detect other adulterants of milk and adulterated food.

针对乳制品掺假日益严重的现象，国家食品监督局、乳制品企业、消费者都亟需一种能快速对复杂牛奶体系中掺杂痕量的目标物进行检测的方法。二维相关谱是解决常规二维谱所遇三大问题（选择性低，提取特征谱信息难，图谱解析难）的一条有效途径。本项目选择牛奶中典型掺杂物三聚氰胺和尿素为目标物，分别以掺杂物浓度、温度为外扰，构建二维相关谱（IR-IR,NIR-NIR,IR-NIR），研究直接对二维相关谱平滑和标准化的预处理方法；研究两种重构二维相关谱深层次挖掘有意义特征光谱信息的数学方法，建立"特征指纹图谱库"，提取相关谱矩阵特征参数，建立"特征向量库"；在此基础上，建立特征峰比对、特征向量结合模式识别两种掺伪牛奶的判定方法；以二维及多维偏最小二乘法为基础，选择随外扰变化敏感的特征光谱信息区域为建模区间，建立定量分析掺伪牛奶的数学模型，并进行优化。该方法可以拓展到牛奶中其它掺杂物的检测以及其它食品掺伪检测中。

近年来，牛奶质量与安全问题日益严重，因此寻求一种快速、廉价、精确的牛奶质量检测方法已成为当今乳制品行业急需解决的重大问题之一。二维相关谱技术以其高光谱分辨率、高选择性和高图谱解析能力，可有效解决常规一维谱在数据分析过程中面临的三类困难：①光谱选择性低；②特征信息提取难；③图谱解析难，特别适合于那些传统一维光谱方法难以满足的相似样品的判别分析。. 本项目以常见的掺杂牛奶为研究对象，如掺杂尿素牛奶和掺杂三聚氰胺牛奶，明确了其二维相关(IR, NIR 和IR/NIR)谱特性；研究了不同标准化方法对不同复杂体系二维相关谱的影响；探讨了重构二维相关谱以深层次挖掘微弱特征信息的数学方法；提出了一种揭示微弱特征峰的解析方法：依据同步二维相关谱交叉峰的正负，并结合异步谱交叉峰的有无来实现掺杂物官能团来源的确认；发展了表观统计二维相关谱参数化和多维主成分二维相关谱参量化方法，并与模式识别结合实现掺杂牛奶的判定，该方法进一步提取二维相关谱特征信息，缩短了计算时间，提高了建模效率；提出了基于非线性核隐变量正交投影算法和二维相关谱结合判别掺杂牛奶的方法，该方法有效解决了掺杂牛奶体系的非线性问题，提高了预测能力；构建基于同步-异步二维相关谱掺杂牛奶的判别模型，与同步和异步二维相关谱模型相比，该模型既包含了同步谱中“相似性”信息，又包含了异步谱中“差异性”信息，同时剔除了冗余的信息，从而提高了掺杂牛奶的判别正确率；发展了一种基于异谱二维IR/NIR相关谱掺杂牛奶判别方法，该方法可进一步放大和提取牛奶中掺杂物的特征信息，因此具有较好的预测能力；建立了基于二维相关谱定量分析牛奶中掺杂物的多维偏最小二乘模型，并与一维谱的偏最小二乘模型相比，该方法可提高掺杂牛奶的预测精度；此外，研究了参考谱选择、中心化、标准化等因素对二维相关谱模型的影响。. 本项目主要在提取牛奶中掺杂物特征信息的基础上，将二维相关谱与化学计量学方法结合起来实现掺杂牛奶的检测。项目所取得的成果，为进一步推广该方法在其它食品安全领域中的应用提供理论和实验依据。.