基于比色光谱的蔬菜有机磷农药残留快速检测方法研究

发布时间：2018-04-11 13:45

本文选题：比色光谱 + 含硫有机磷　；参考：《中国农业大学》2016年博士论文

【摘要】：有机磷农药的快速检测是保证农产品品质安全的重要环节。本论文使用比色光谱技术开展了蔬菜中典型有机磷农药检测研究,为农药快速检测设备的研制提供理论基础和技术保障。主要研究内容如下：[1]典型有机磷农药残留快速检测比色试剂优选研究经过大量试验,对氯化钯比色法进一步改进和优化,分别使用乙酸和氯化钠代替传统的浓盐酸作为氯化钯的溶剂,对含硫基类有机磷农药乐果、毒死蜱、氧乐果和乙酰甲胺磷试验,比色反应后的吸收光谱可以检测的最大残留量范围在0.05～0.5 mg/kg,满足或基本满足国标GB2763-2014规定的部分蔬菜的检测要求。室温下比色反应时间为2 min,满足了快速、安全、简便的检测要求。[2]含硫基类有机磷农残预测模型研究根据4种农药的比色光谱,分别在最大吸收波长处建立一元线性回归模型,但预测精度普遍较低。分析原因是比色反应的影响因素较多,实验误差很难完全消除。为此,引入PCA、PLS、BP/RBF:神经网络等方法建立预测模型。试验表明PLS回归效果较好。以乐果为例,在400-600nm波段,主成分为4时,训练集R2=0.9941,RMSEP=2.7703,验证集R2=0.9933,RMSEP=2.2148,表明利用可见光波段检测农药残留的可行性。[3]提高定量预测普适性的分段多模型方法研究在实际检测中,由于待检物中农药含量无法预测,导致无法加入适量的比色试剂。在比色试剂严重过量或不足时,比色曲线会发生较大变化,预测模型的建立变得复杂。为此,在使用PLS、PCA+BP/RBF神经网络等方法建立单一模型的基础上,提出了“分段多模型”的建模策略,即先使用SVM方法区分待检物的浓度段,然后在每个浓度段内使用PLS模型定量预测。试验表明：SVM+PLS方法在待检农药浓度较大时明显提高了预测精度,具有较强的普适性。[4]有机磷农药种类快速识别模型研究根据不同农药的氯化钯比色光谱特征建立了农药种类识别模型。对比PCA+BP/RBF神经网络、CV+SVM、PSO+SVM方法,PSO+SVM方法取得了较好的效果。在4种农药的分类识别中,优化参数c=2.6857,g=0.01时,训练集和验证集的识别准确率分别达到100%和95%,为识别农药种类提供了一种快速、可行的新方法。[5]蔬菜基质对有机磷农药检测的影响分析及消除方法研究了不同蔬菜基质下有机磷农药比色光谱的变化,在吸光值和谱线变化趋势两方面对比了与纯农药光谱的差异。分析并得出结论：不同蔬菜对不同有机磷农药比色光谱的影响波段不同,在影响波段吸光值的区分度明显降低。为此,使用多浓度段差示法采集光谱数据,提高了模型预测精度,对消除蔬菜影响有一定效果。本研究在大量实验的基础上优化了有机磷农药的比色试剂；建立了普适性较强的含硫有机磷农药预测模型；分析了常用蔬菜对农药检测的影响及改进方法,最终确定了可行的预测模型。
[Abstract]:Rapid detection of organophosphorus pesticides is an important step to ensure the quality and safety of agricultural products.In this paper, the detection of typical organophosphorus pesticides in vegetables was carried out by colorimetric spectroscopy, which provided the theoretical basis and technical guarantee for the development of rapid detection equipment for pesticides.The main contents of this study are as follows: [1] the rapid detection of organophosphorus pesticide residues by colorimetric reagent was studied. After a large number of experiments, the palladium chloride colorimetric method was further improved and optimized.Acetic acid and sodium chloride were used instead of the traditional concentrated hydrochloric acid as palladium chloride solvent to test dimethoate, chlorpyrifos, oxymethamidophos and oxymethamidophos.The maximum residue can be detected in the range of 0. 05 ~ 0. 5 mg 路kg ~ (-1) mg 路kg ~ (-1) by colorimetric reaction, which meets or basically meets the requirements of some vegetables determined by GB2763-2014.The colorimetric reaction time was 2 min at room temperature, which met the requirements of rapid, safe and simple detection. [2] the prediction model of organic phosphorus residues containing sulfur groups was studied. According to the colorimetric spectra of four pesticides, a linear regression model was established at the maximum absorption wavelength, respectively.However, the prediction accuracy is generally low.The reason is that there are many factors affecting the colorimetric reaction, and the experimental error is difficult to be completely eliminated.For this reason, the prediction model is established by using PCAA PLSU BP / RBF (neural network) and so on.The results showed that PLS regression was effective.Taking dimethoate as an example, in the 400-600nm band, when the principal component is 4, the training set R2X 0.9941 RMS EPN 2.7703 is verified, which shows the feasibility of using visible light wave band to detect pesticide residues. [3] the segmented multi-model method to improve the universality of quantitative prediction is studied in practice.Due to the unpredictable content of pesticides in the sample, it is impossible to add a suitable amount of colorimetric reagent.When the colorimetric reagent is seriously excessive or insufficient, the colorimetric curve will change greatly, and the establishment of prediction model becomes more complicated.Therefore, on the basis of establishing a single model by using BP/RBF neural network and other methods, a "segmental multi-model" modeling strategy is proposed, that is to say, the method of SVM is used to distinguish the concentration segment of the object to be detected.Then the PLS model was used for quantitative prediction in each concentration range.The experimental results show that the prediction accuracy of the PLS method is improved obviously when the pesticide concentration is high.[4] the rapid recognition model of organophosphorus pesticides was established according to the palladium chloride spectral characteristics of different pesticides.Compared with the PCA BP/RBF neural network, the CV SVM SVM method and the PSO SVM method have achieved good results.In the classification and recognition of 4 kinds of pesticides, the accuracy of training set and verification set were 100% and 95%, respectively, when the optimized parameter was 2.6857 g / g ~ (0.01), which provided a fast method for the identification of pesticide species.The effect of vegetable substrates on the determination of organophosphorus pesticides and its elimination methods were studied, and the changes of the colorimetric spectra of organophosphorus pesticides under different vegetable substrates were studied.The difference between absorption value and spectral line change trend was compared with that of pure pesticide.It is concluded that the influence of different vegetables on the colorimetric spectrum of different organophosphorus pesticides is different in different bands, and the differentiation of absorptivity in the affected bands is obviously reduced.Therefore, the spectral data were collected by multi-concentration section difference method, which improved the precision of model prediction and had certain effect on eliminating the influence of vegetables.On the basis of a large number of experiments, the colorimetric reagent of organophosphorus pesticide was optimized, the prediction model of sulfur containing organophosphorus pesticide was established, the influence of common vegetables on pesticide detection and the improved method were analyzed.Finally, the feasible prediction model is determined.
【学位授予单位】：中国农业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：S481.8

【相似文献】