乳制品安全拉曼光谱成像分析新方法研究

发布时间：2018-03-26 08:09

本文选题：拉曼光谱成像　切入点：奶粉掺杂检测　出处：《天津大学》2016年硕士论文

【摘要】：乳制品作为人类的重要食品来源之一,其消费量十分巨大。近年来,不法分子在经济利益的驱动下,以次充好并大肆造假,严重危及消费者生命健康。如何高效识别并检测奶粉中的掺假物质,成为当前乳制品安全领域的研究热点,在我国当前的国情下具有至关重要的意义。奶粉是一种复杂多组分的非均匀固体粉末,由于掺假物范围很广且浓度分布不均,在奶粉中检测低浓度的掺假物,对分析技术提出了很高的要求,既需要高效的微区分析技术,又需要快速的大面积扫描技术。在各种检测技术中,拉曼光谱成像技术是为数不多能同时满足微区分析和大面积扫描的快速检测技术之一,并以其特有的高分辨、高通量等优点,在奶粉的掺假物质分析中具备较大的应用潜力。因此,本论文尝试采用拉曼光谱成像技术对奶粉体系进行掺假检测。在拉曼光谱成像检测过程中,其成像信号尚存在重叠严重、背景干扰强等问题。因此,本论文针对不同奶粉掺假物质所带来的额外波动性,发展了一种新型数据驱动的多尺度建模方法(Data-Driven Multiscale Modeling,DDMM)。该算法将自适应调整拉曼信号的小波分解函数,使其局部动态地逼近信号本征信息,进而最大限度地克服样本的波动性,以避免模型失效。在此基础上,本论文还发展了一种基于最小二乘支持向量回归的非线性算法(Modified discrete wavelet transform Elimination of uninformative variables Least square support vector regression,MDWT-UVE-LSSVR),用于复杂体系分析。该算法有机结合了改进小波变换、无用信息变量删除法、及最小二乘支持向量回归,用于分别扣除复杂光谱体系中的荧光背景、基质及非线性干扰。该算法被应用于奶粉掺假检测,结果表明,该算法能够实现的对不同掺杂物质的判别与区分,模型总体识别正确率为99.74%,并且进行半定量分析效果。本论文尝试利用拉曼光谱成像技术对奶粉掺假体系进行快速检测,并发展了DDMM和MDWT-UVE-LSSVR算法对其成像信息进行准确解析。结果表明,复杂体系的光谱解析往往需要有效结合不同的算法以克服不同来源的干扰。而本论文发展的这两种算法可以有效克服不同掺假物质带来的额外波动信息,显著提升拉曼光谱成像技术预测的稳定性和预测精度,并且无需对光谱数据结构进行预定假设,为奶粉掺假快速检测提供了一种有力工具。
[Abstract]:Dairy products are one of the most important food sources for human beings, and their consumption is very great. How to efficiently identify and detect adulterated substances in milk powder has become a hot research topic in the field of dairy products safety. Milk powder is a kind of complex multicomponent nonuniform solid powder. Because of the wide range of adulterants and uneven concentration distribution, milk powder is used to detect low concentration adulterates in milk powder. High requirements are put forward for analytical techniques, which require both efficient micro-area analysis technology and rapid large area scanning technology. In all kinds of detection technologies, Raman spectroscopy imaging is one of the few rapid detection techniques that can satisfy both microanalysis and large area scanning, and has the advantages of high resolution and high throughput. Therefore, this paper attempts to use Raman spectroscopy imaging technology to detect adulteration of milk powder system. In the process of Raman spectrum imaging detection, the imaging signals of milk powder are still overlapped seriously. Therefore, this paper aims at the additional volatility caused by different milk powder adulterated substances. A new data-driven multi-scale modeling method, Data-Driven Multiscale Modeling, is developed. The algorithm adaptively adjusts the wavelet decomposition function of the Raman signal to make it approach the intrinsic signal information dynamically and overcome the volatility of the sample to the maximum extent. To avoid model failure. On this basis, This paper also develops a nonlinear algorithm based on least squares support vector regression (LS-SVM), which is modified discrete wavelet transform Elimination of uninformative variables Least square support vector regression MDWT-UVE-LSS VRV for complex system analysis. And least squares support vector regression (LSVR), which is used to deduct fluorescence background, matrix and nonlinear interference from complex spectral system respectively. The algorithm is applied to the detection of milk powder adulteration. The results show that, The recognition accuracy of the model is 99.74, and the effect of semi-quantitative analysis is achieved. In this paper, Raman spectroscopy imaging is used to detect the adulteration system of milk powder. DDMM and MDWT-UVE-LSSVR algorithms are developed to accurately analyze the imaging information. The results show that, Spectral analysis of complex systems often requires effective combination of different algorithms to overcome interference from different sources, and the two algorithms developed in this paper can effectively overcome the additional wave information brought by different adulterated substances. The prediction stability and prediction accuracy of Raman spectroscopy imaging technology are improved significantly, and the spectral data structure is not presupposed, which provides a powerful tool for rapid detection of milk powder adulteration.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TS252.7;O657.37

【相似文献】