【作者】 王晓燕；

【导师】 陈仁文；

【作者基本信息】 南京航空航天大学 ， 测试计量技术及仪器， 2020， 博士

【摘要】 农药残留超标是影响食品安全的重要因素,快速有效的农药残留检测方法以及安全有效的农药残留降解技术研究对保障食品安全具有重要意义。本文以果汁中的农药残留为研究对象,首先,应用光谱法检测农药残留并采用线性方法对低浓度下的农药残留进行了分析;然后,研究了高浓度下农药残留的非线性回归方法,并分别建立了农药种类的分类识别模型以及多组分混合农药残留的多回归模型,实现了对残留农药的分类及其浓度的准确估计。最后,研究了紫外光、臭氧等农药残留降解方法,并对降解效果进行了表征。论文主要工作如下:(1)基于单特征峰的低浓度单组分农药残留检测与分析。根据朗伯-比尔定律,在低浓度溶液中,荧光强度与溶液浓度基本呈线性关系,而在高浓度时,由于猝灭及自吸等现象使得两者之间不再呈线性关系。对于低浓度下的农药残留,分别应用荧光检测、吸收检测方法研究了灭蝇胺、异丙甲草胺在不同果汁中的光谱特性,并根据其荧光或吸收特征峰值建立了农药浓度线性模型,获得了检测限、定量限、线性范围及回收率参数,并比较分析了两种检测方法的性能差异以及背景果汁不同时对检测性能的影响。针对异丙甲草胺浓度过低时其吸收峰不突出的情况,对吸收光谱进行求导后重新建模分析,得到了较尖锐的特征峰并提高了检测性能。实验表明以上方法可对低浓度农药残留进行有效检测。(2)基于多特征波长点的高浓度单组分农药残留检测方法研究。以苹果汁中克菌丹残留的荧光检测为例,共配制了低浓度以及超过线性范围高浓度下的153组实验样本。针对特征峰荧光强度与高浓度克菌丹不再符合线性关系的特点,分别采用多个特征波长点和全光谱的方法对光谱特性进行了分析。首先分别应用遗传算法、粒子群算法、基于二次特征选择的改进混合粒子群算法、以及连续投影等算法,从光谱中优选出特征波长。在此基础上,分别采用最小二乘、偏最小二乘、主成分分析和支持向量机四种方法,建立了农药残留浓度的线性及非线性回归模型,并通过实验对以上模型进行了验证,确定了不同情况下模型的适用性。(3)残留农药的种类识别与多组分混合农药的检测方法研究。以灭蝇胺、异丙甲草胺、克菌丹、噻虫嗪四种农药残留为研究对象,应用支持向量机、最小二乘支持向量机以及主成分分析方法,对四种组分进行了分类;以上述四种农药残留的混合溶液为研究对象,建立了全光谱和特征光谱两种模型对各农药浓度进行估计。在全光谱分析中发现,最小二乘支持向量机比支持向量机具有更好的性能。应用连续投影算法优选出特征波长,采用性能较优的最小二乘支持向量机建立了特征光谱模型。与偏最小二乘算法进行了性能对比,结果表明,采用特征光谱的最小二乘支持向量机模型最优。(4)农药残留降解效果评估。在农药残留检测的基础上,分别应用紫外光、臭氧方法对果汁中农药残留进行了降解实验。应用荧光光谱方法研究了果汁中农药残留的降解特性。利用其荧光光谱特征峰的强度变化表征降解过程,建立了降解率与降解作用时间之间的数学关系式,评估了降解效果,并明确了不同农药适合的最佳降解方法。论文应用光谱法检测农药残留,基于荧光光谱特性表征降解过程并建立降解模型,为深入了解降解规律提供了可靠方法和思路;基于非线性回归的分析方法大大提高了高浓度下农药残留含量的预测精度;在优选特征波长时,提出的基于二次特征选择改进混合粒子群算法可以有效筛选出特征光谱;结合支持向量机算法建立的农药含量预测模型,收敛速度快且预测性能良好,为实现高浓度农药残留的精确定量分析提供了参考方法;农药种类识别研究为快速实现农药种类鉴别提供了最佳建模方法,而多组分下的回归分析则实现了光谱重叠背景下混合溶液中各农药的精确浓度估计。更多还原

【Abstract】 Excessive pesticide residues are one of the important factors that endanger food safety,thus,research of rapid detection methods and effective degradation technologies of pesticide residues are of great significance for food safety.This dissertation focuses on fruit juice pesticide residues detection technologies.Firstly,the pesticide residues are detected by spectrophotometry and analyzed with linear method at low concentration.Then,a non-linear regression method for pesticide residues under high concentration is put forward,the recognition and regression models of mixed pesticide residues are established respectively,resulting in good accurate prediction of classification and concentration of pesticide residues.Finally,the degradation methods of pesticide residues using ultraviolet and ozone are studied,and the degradation effect is characterized and evaluated.The main contributions of this dissertation are as follows:(1)Detection and analysis of single component low concentration pesticide residues based on single characteristic peak are studied.According to Lambert-Beer Law,at low concentration solution,the fluorescence intensity and the solution concentration are basically linear,while at high concentration,there is no longer a linear relationship between them due to quenching and self-absorption.For the pesticide residues at low concentration,the spectral characteristics of cyromazine and metolachlor in different fruit juices are studied by utilizing fluorescence detection and absorption detection,and the linear model of pesticide concentration is established according to the peak value of fluorescence or absorption characteristics.The parameters such as limit of detection,limit of quantitation,linear range and recovery rate are obtained,and the performances between the two detection methods,as well as the influence of the two different basic juices on detection performance are compared.In view of the fact that the absorption peak of metolachlor is not prominent at low concentration,the absorption spectrum is derived for further analysis.The sharp characteristic peak is obtained and the detection performance is improved.The experimental results show that the above methods can effectively detect the pesticide residues at low concentration.(2)Detection methods of single component pesticide residues at high concentration are studied based on multi characteristic wavelength points.Taking the captan pesticide residues in apple juice as an example,153 groups of samples at low concentration and high concentration beyond the linear range were prepared.Regarding the fact that the fluorescence intensity of the characteristic peak is no longer linear with the captan concentration,characteristic wavelength points and full spectrum are used to analyze the spectral characteristics.Firstly,four analysis methods including genetic algorithm,particle swarm optimization algorithm,improved hybrid particle swarm optimization algorithm based on secondary feature selection,and successive projections algorithm are applied respectively to choose the characteristic wavelengths from the spectrum.Then,the linear and non-linear regression models of concentration prediction for pesticide residues are established using least square,partial least square,principal component analysis and support vector machine methods.The above models are verified by experiments,and the model applicability in different situations is determined.(3)The identification modeling of pesticide types and regression modeling of multi-component residues are researched.Support vector machine classifier,least square support vector machine classifier and principal component analysis algorithms are applied to realize the identification of four kinds of pesticide residues including cyromazine,metolachlor,captan and thiamethoxam.Full spectrum model and feature spectrum model are established for the regression of four mixture pesticide residues.It is found that least square support vector machine regression had better performance than support vector machine regression in full spectrum analysis.The successive projections algorithm is used to optimize the characteristic wavelength,and the least square support vector machine regression with better performance is used to establish the characteristic spectrum model.Compared with the partial least square algorithm,the results show that the least square support vector machine model with characteristic spectrum is better.(4)Degradation effect of pesticide residues is evaluated.Based on the detection of pesticide residues,the degradation experiments of pesticide residues in fruit juice are carried out using ultraviolet and ozone methods,and the degradation characteristics are analyzed by fluorescence spectroscopy.The degradation effect is characterized by the change of the fluorescence intensity,the mathematical relationship between degradation rate and degradation time is deduced as well.Thus,the degradation effect is evaluated and the best degradation method for different pesticides is determined at last.It is found that,in this dissertation,the degradation model based on fluorescence spectral characteristics can provide a reliable approach for further understanding of degradation rules.Nonlinear regression greatly improves the prediction accuracy of pesticide residues for high concentration samples.The proposed improved hybrid particle swarm optimization method can effectively achieve the effective wavelength selection.The prediction model based on support vector machine has fast convergence speed and good prediction performance,which provides a reference method for accurate prediction of pesticide residues.The identification research provides the best method for the rapid identification of pesticide types,and the multi-component regression analysis has achieved high accuracy pesticide concentration prediction of each pesticide in mixed solution under overlapping spectrum.更多还原