基于近红外光谱技术的蔬菜农药残留种类检测

该文基于近红外光谱技术，提出一种快速无损检测方法，以期实现蔬菜农药残留的分类检测。通过对喷洒了氰戊菊酯溶液、三唑磷溶液和未喷洒农药的生菜样本进行研究，比较不同预处理后的建模效果，选用SNV算法作为最优预处理方法。分别采用连续投影算法（SPA）、自主软收缩法（BOSS）和竞争性自适应重加权算法（CARS）对预处理后的光谱数据进行特征波段选择。采用支持向量机（SVM）和基于灰狼算法（GWO）优化的支持向量机（SVM）算法对特征波长变量分别建立分类模型。再通过对建立的模型进行比较得出:CARS-GWO-SVM模型取得了最佳的分类效果，模型的训练集精度和预测集精度均为100%。因此，利用近红外光谱技术对蔬菜上的农药残留进行分类检测是可行的。该研究为生菜中其他农药残留的快速无损检测分析提供参考。 

Detection of Pesticide Residues in Vegetables Based on Near-infrared Spectroscopy

Based on near-infrared spectroscopy，a non-destructive testing method to achieve rapid and non-destructive classification of pesticide residues on vegetables was proposed.It hope to realize the classification and detection of pesticide residues on vegetables.By studing the lettuce samples sprayed with cypermethrin solution，triazophos solution and unsprayed pesticide，the modeling effects of different pretreatments were compared，and SNV algorithm was chosen as the optimal pre-processing method for this study.In this study，the successive projection algorithm（SPA），the bootstrapping soft shrinkage（BOSS）and the competitive adaptive reweighted sampling（CARS）were used to select the characteristic bands of the preprocessed spectral data.A support vector machine（SVM）and the support vector machine（SVM）algorithm based on Grey wolf optimization（GWO）optimization were used to establish a classification model of characteristic wavelength variables.The results showed that the CARS-GWO-SVM model achieved the best classification influence，and the accuracy of training set and the accuracy of prediction set were both 100%.Therefore，it is feasible to use NIR spectroscopy to classify pesticide residues on vegetables.This study also provided a reference for the rapid，non-destructive analysis of other pesticide residues in lettuce.