【作者】 陈国庆；

【导师】 朱拓；

【作者基本信息】 江南大学 ， 控制理论与控制工程， 2010， 博士

【摘要】 本文应用荧光光谱结合人工神经网络技术鉴别和测定了合成食品色素、中国白酒、黄酒和葡萄酒等发酵酒,以及三聚氰胺等与食品安全相关的物质,并讨论了这些荧光光谱的特性,进而对食品安全实时荧光监控系统进行了基础设计。实验测得胭脂红、苋菜红、诱惑红、酸性红、赤藓红、新红、日落黄、柠檬黄、喹啉黄、亮蓝和靛蓝等11种目前我国允许使用的合成食品色素和常被非法用于食品的工业色素苏丹红Ⅰ和苏丹红Ⅳ的三维荧光光谱,以及这些色素10种浓度的溶液在各自最佳激发波长光激发下的荧光发射光谱,得到了各自的荧光光谱特性参数和荧光强度与溶液浓度的关系。结果表明,这些色素在短波长光激发下,均能产生强的荧光,各自的荧光光谱具有明确的特征。提出并应用全部导入、小波变换、逐阶求导等新方法提取和处理荧光光谱特征数据,利用这些数据训练和建立了BP、RBF和PNN等神经网络,实现了对合成色素方便、快捷、准确的种类鉴别和浓度测定,鉴别准确率均达到100%,测定平均相对误差均在4%以内。首次实现了工业色素和食品色素的准确识别,以及混合色素中各色素浓度的同时测定。实验测得新的白酒样品的荧光光谱,建立了中国白酒荧光光谱图库,首次确定以荧光谱峰个数、荧光峰值波长、荧光最佳激发波长、荧光发射谱1/4/、1/2、3/4处线宽6个参数作为特征数据并建立特征向量,以其欧氏距离作为鉴别判据,提出并应用自动提取、搜索比对、判别归类、向量运算、阈值判定等原理和方法,发挥了荧光光谱技术和计算机智能技术的优点,建立了基于荧光光谱的中国白酒鉴别系统,实现了中国白酒品种和白酒年份酒年份的科学化、仪器化和智能化鉴别。首次将荧光光谱技术应用于三聚氰胺的定量检测,测定三聚氰胺浓度的平均相对误差为1.50%。检测并分析了食用油及其加热后的荧光光谱,结果表明这种食用油在波长为400nm左右的光激发下,能产生较强的荧光。得到了在不同激发光下两个荧光峰随加热次数和加热时间的变化规律,对其成因进行了分析。实验研究了农药以及黄酒、啤酒和葡萄酒3种发酵酒的荧光光谱,结果表明,荧光光谱是荧光物质的指纹图谱,荧光光谱特性是荧光物质的特征信息,可用于对物质进行定性和定量检测。研究了食品安全实时荧光光谱检测和监控系统,提出了系统的设计方案,以及系统结构、组成及配置。本文的研究结果为丰富和发展食品安全检测技术提供新思路,可为促进食品安全监管领域的方法创新和进步提供技术支持。更多还原

【Abstract】 Using fluorescence spectroscopy and artificial neural network method, we made successful measurement and identification of some materials in this letter, such as synthetic food colors, chinese liquors, fermented wines including rice wines and wines, stuffs relating to food safety as of melamine, and et al. Characteristics and features of these fluorescence spectra are discussed, and basic systems for real time monitoring of food safety using fluorescence spectroscopy are further designed.In the experiments, we successfully obtained the three-dimensional fluorescence spectra of eleven synthetic food colors (including Ponceau 4R, Amaranth, Allura Red, Acid Red, Erythrosine, New Red, Sunset Yellow, Tartrazine, Quinoline Yellow, Brilliant Blue, and Indigotine) allowed by-Chinese government currently and several industrial pigments (including Sudan I and Sudan IV) which are often added to food illegally. Fluorescence emission spectra of these synthetic food colors and industrial pigments are also obtained under the conditions that the solutions are detected at ten different concentrations with their own optimal excitation wavelength. With these experiments above, spectral features of each synthetic food color or industrial pigment can be achieved. The relationship between fluorescence intensity and the corresponding concentration of the solution can also be determined. The results indicate that with excitation of short wavelength these synthetic food colors and industrial pigments all produce fluorescence light of high intensity, and each fluorescence spectra of them has its distinct characteristics.After that we extracted some characteristic parameters from the fluorescence spectra and handled them by some new methods such as the wavelet transformation and derivative spectra. Using the data above, we built up and trained the BP, RBF and PNN neural networks through which we can not only identify these synthetic food colors easily, quickly and precisely but also determine the concentration of their solutions. The accuracy rates of identification reach 100%, and the average relative error remains under 4% for each detection:Here, for the first time, we not only accurately identified these synthetic food colors and industrial pigments but also determined the concentration of them in mixed solutions at the same time.From the experiments we built up the fluorescence data base of Chinese liquors and added some new data from fluorescence data of recent liquor samples. We initially convert six characteristic parameters to one characteristic vector. The six parameters include numbers of fluorescence peaks, wavelength at maximum fluoresence intensity, optimal excitation wavelength, and three linewidths at 1/4,1/2,3/4 of maximum fluorescence emission spectra. Based on Euclidean distance, using methods such as automatic extraction, search and comparison, discriminant classification, vector operation, and decision of threshold values, we built the identification system of Chinese liquors on fluorescence spectra. The system exploits the advantage of fluorescence spectroscopy and computer intelligent technology, and realizes the identification of Chinese liquors from their species and years in a scientific, instrumental, and intelligent manner.In addition, we made quantitative measurement of concentrations of melamine using fluorescence spectroscopy technique for the first time with an average relative error under 1.5%. We also detected and analyzed the fluorescence data of the cooking oil after it was heated. The result shows that under light excitation of 400 nm the cooking oil produces fluorescent light with high intensity. We even obtained the rules how the two fluorescence peaks change with heating time and the number of times, and made some analysis about them. Moreover, we studied the fluorescence spectroscopy of some pesticides and some fermented wines including rice wines, beers, and wines. All the experimental results above indicate that fluorescence spectroscopy which can be used in both qualitative and quantitative detection, is the fingerprint and the characteristic information of the fluorescent materials.We further made a feasibility study of setting up a system for real time monitoring of food safety using fluorescence spectroscopy. We proposed some design solutions, and details about elements, structures and configurations were carefully considered.In conclusion, the work in this letter not only provides a new way in riching and developing the detecting techniques for food safety, but also provides technical support in innovative methods and improvements on food safety supervision.更多还原