基于近红外/荧光/偏振多源光谱信息融合的蔬菜有机磷农药残留高精度检测机理及方法研究

The detection of pesticide residues is the premise of vegetable food safety. In this project, lettuce will be chosen as carrier of pesticide residues.In different varieties of pesticide residues, the relationship between the organic phosphorus pesticide group containing phosphorus and near-infrared spectra,and the the relationship between organic phosphorus molecular structure and fluorescence spectra will be studied, and the identification mechanism of pesticide residues' category will be researched. In the different concentrations of organic phosphorus pesticide,the change of infrared spectroscopy caused by the change of the organic phosphorus pesticide group containing phosphorus, and the change of fluorescence spectroscopy caused by organic phosphorus molecular structure will be studied.The effect of the change of micro structure such as leaf sponge and fence to polarization characteristic and polarization degree, and the relationship between pesticide residues and polarization direction and polarization distribution will be analysized.The nondestructive detection mechanism of pesticide residues based on near-infrared spectra,fluorescence spectra and polarization spectra. The characteristic wavelengths of near-infrared spectra and fluorescence spectra will be selected, and the category of pesticide residues will be identified.The features of near infrared spectra,fluorescence spectra and polarization spectra will be extracted, and feature space will be optimizated. The precide detection model will be structed based on information fusion technology and modern mathematical regression algorithm,and verification test will be made.This project aims to put forward a kind of precide nondestructive method to detect pesticide residues in vegetable based on multi-source information of nearinfrared spectra, fluorescence spectra and polarization spectra. This method breaks through the shortage of traditional single spectral method ,such as low precision and poor universality . It will provide a basis for safety usage of vegetable.

农药残留检测是蔬菜食品安全的保障与前提。本项目以生菜为载体，研究不同品种农药残留下，有机磷农药含磷基团与近红外光谱的关系及有机磷大分子结构与荧光光谱的关系，探讨农残类别鉴别机理；在不同农残水平下，研究有机磷农药含磷基团变化引起近红外光谱变化规律，有机磷大分子结构引起荧光光谱变化规律，研究叶片海绵体和栅栏等微结构特征变化对偏振特性和偏振度的影响，分析农残与偏振特性和偏振度分布的对应关系，探索基于近红外/荧光/偏振多源光谱的农残量无损检测机理。确定各有机磷农残的有效近红外光谱敏感波段和荧光光谱特征，进行农残类别鉴别。提取有效近红外光谱、荧光光谱及偏振光谱特征，优化特征空间，利用信息融合技术及现代数学回归算法构建生菜叶片农药残留高精度检测模型，进行验证试验。本项目旨在提出一种基于近红外/荧光/偏振多源光谱信息融合的蔬菜农残高精度无损检测方法，突破了传统单一光谱方法检测普适性差、精度低的局限性。

本项目以生菜为载体，无土栽培生菜并喷洒不同种类、水平的有机磷农药，观察作物样本叶片表面的微结构、内部组织结构变化。通过相关电镜实验，结果表明，随着喷洒的乐果农药浓度的增加，生菜叶片表面气孔的长宽比和气孔密度也将逐渐变小，生菜叶片厚度变薄，生菜叶片内部的嗜锇颗粒增加，淀粉颗粒将变少，叶绿体间间隙逐渐变大。叶片细胞的排列结构方式将直接影响植被反射光谱。喷洒不同浓度乐果农药的生菜叶片光谱信息间的差异性进一步验证了光谱技术应用于生菜农药残留检测的有效性。研究了近红外高光谱图像的获取与处理，结果表明，从4个角度（0°、45°、90°、135°）对PC1和PC2图像进行基于灰度共生矩阵的纹理特征（对比度、相关性、能量、同质性）提取，结合连续投影算法提取得到的特征光谱所建立的Fisher模型较好。研究了荧光光谱的预处理算法，分析比较了多种预处理算法，发现SG-SNV detrending预处理算法最好。研究了光谱信息降维算法，采用小波变换分解得到信号细节组分和信号近似组分，通过对信号细节组分（高频部分）的奇异值分解，提取较大奇异值对应波长作为特征波长。项目组在离散小波变换（DWT）特征提取算法基础上，结合近红外光区主要含氢原子团（C-H、N-H、O-H）伸缩振动的倍频及组合频中心谱区，提出一种分段离散小波变换（PDWT）特征提取算法。此外，项目组提出了一种小波变换（WT）与MD-MCCV相结合的算法WT-MD-MCCV。偏振光谱定性检测农药残留，最佳入射天顶角A，探测天顶角B，光源偏振角C，探测器偏振角D和样品台方位角E参数组合分别为60°、45°、0°、30°和270°。基于近红外/荧光/偏振多源光谱信息融合的生菜叶片农药残留建模分析研究结果表明，不同的光谱数据所建立的SVR模型性能也存在差异，多源光谱信息特征所建立的SVR模型性能要优于单一光谱（近红外光谱、荧光光谱、偏振光谱）所建立的SVR模型。项目组提出了这种基于多源光谱信息融合的生菜农药残留无损检测方法，突破了传统单一光谱方法检测普适性差、精度低的局限性。