基于荧光光谱的白菜混合农药残留联合含量检测

Detection of Mixed Pesticide Residues in Cabbage Based on Fluorescence Spectroscopy

摘要:

苯甲·吡唑醚是农作物生产种植中常用的杀菌剂。为了高效监控蔬菜中农药残留情况，本研究利用荧光光谱技术检测白菜中苯甲·吡唑醚混合农药残留含量。首先通过三维荧光光谱确定470 nm为苯甲·吡唑醚的最佳激发波长;其次分别以吡唑醚菌酯和苯醚甲环唑含量为对象，利用随机跳蛙(Random frog, RF)、遗传算法(Genetic algorithm, GA)、变量迭代空间收缩算法(Variable iterative space shrinkage algorithm, VISSA)进行特征筛选;卷积神经网络(Convolutional neural network, CNN)被用于荧光光谱建模，同时与反向传播神经网络(Backward propagation neural network, BP)和极限学习机(Extreme learning machine, ELM)模型作比较。结果显示VISSA为最佳特征波长选择方法，CNN模型获得了苯甲·吡唑醚含量最佳预测效果，吡唑醚菌酯和苯醚甲环唑同时预测的测试集决定系数均达0.967，均方根误差分别达3.455、5.760 mg/kg。试验结果表明，荧光光谱技术结合深度学习算法同时预测混合农残含量的可行性，可为在线检测农残含量系统的开发提供理论依据。

Abstract:

Benzyl-pyrazole is a popular fungicide used in crop production and cultivation. In order to efficiently monitor the pesticide residues in vegetables, fluorescence spectroscopy was used to detect the content of mixed benzyl-pyrazole ether pesticide residues in cabbage. Firstly, 470 nm was determined as the optimal excitation wavelength of benzyl-pyrazole ether by three-dimensional fluorescence spectroscopy. In addition, random frog (RF), genetic algorithm (GA) and variable iterative space shrinkage algorithm (VISSA) were used for the feature selection of pyraclostrobine and difenoconazole, respectively. Convolutional neural network (CNN) was used for fluorescence spectroscopy modeling while comparing with backward propagation neural network (BP) and extreme learning machine (ELM) models. The results showed that VISSA was the best feature wavelength selection method, and the CNN model obtained the best prediction for benzyl-pyrazole. The coefficient of determination of the test set for both pyraclostrobine and difenoconazole reached 0.967, and the root mean square error reached 3.455 mg/kg and 5.760 mg/kg, respectively. It was demonstrated the feasibility of fluorescence spectroscopy combined with deep learning algorithms for simultaneous prediction of mixed pesticide residue content, which can provide a theoretical basis for the development of online pesticide residue content detection systems.