【摘要】 蔬菜和水果中的拟除虫菊酯类农药的残留会对人体健康造成危害。建立了表面增强拉曼光谱（SERS）结合深度学习模型快速定量检测菠菜中氯氰菊酯农药的方法。首先，Ag/ZnO纳米花作为SERS基底。然后，对采集到的SERS光谱数据进行增强，并分别对增强后的光谱数据进行Savitzky-Golay(S-G)和均值中心化（MC）预处理，后将2种预处理算法联用。分别对原始增强光谱和3种方法预处理后的增强光谱建立了反向传播（BP）神经网络深度学习模型，用于实际样品菠菜中氯氰菊酯的预测。结果表明，基于原始增强光谱的BP神经网络深度学习模型具有优异的预测性能。其测试集的复相关系数R_P=0.9902，均方根误差RMSEP=0.102，检出限为20μg/L，双尾配对t检验表明，标准方法液相色谱-质谱联用（LC-MS）与该方法之间的检测结果无显著差异。将SERS光谱输入到训练好的模型后，5～10 min即可完成检测。建立了一种基于Ag/ZnO纳米花的SERS方法与BP神经网络模型相结合，用于菠菜中氯氰菊酯农药残留的快速定量检测的方法。更多还原

【Abstract】 The residues of pyrethroid pesticides in vegetables and fruits can cause harm to human health. In this work, a surface-enhanced Raman spectroscopy(SERS) coupled with back propagation(BP) neural network deep learning model was established for rapid detection of cypermethrin(CPM) in spinach. Concretely, Ag/ZnO were applied as SERS substrates. Then, the collected SERS spectra were expanded, and the expanded spectral data were preprocessed by Savitzky-Golay(S-G), mean centralization(MC) and the combination of the two methods. Furthermore, BP neural network models were established for the prediction of CPM in actual samples for the original enhanced spectrum and the pretreated spectrum, respectively, and three traditional machine learning models were established for comparison. It was found that the BP model has the best prediction performance based on the original expanded spectrum, with the prediction set R_P=0. 9902, the root mean square error RMSEP=0. 102 and the limit of detection 20 μg/L. The two-tailed paired t-tests showed that there was no significant difference between the standard method LC-MS and this method. The detection can be finished in 5～10 min. This work established an Ag/ZnO-based SERS method coupled with BP neural network model for the rapid and quantitative detection of CPM residues in spinach.更多还原