可见近红外光谱的甘蓝叶片毒死蜱农药残留定性分析

有机磷农药毒死蜱是目前农业生产中使用最广泛的农药之一，但有机磷农药过度使用导致的农药残留却给自然环境和人类生命健康造成严重威胁，因此，开发一种快捷、准确、经济的毒死蜱农药在农产品表面残留的直接检测方法意义重大。配制4组不同体积浓度（1∶200, 1∶500, 1∶800, 1∶1 000）的毒死蜱农药溶液，对照组为纯净水，分别浸泡甘蓝叶片3 min，每组采集30个叶片样本，5组共计150个样本。采用可见近红外光谱仪获取其谱图信息，然后开展不同浓度毒死蜱农药在甘蓝叶片上残留的可见近红外光谱定性分析研究。建模时，将每组数据中24个样本，5组共计120个样本作为建模训练集，剩下每组6个样本，5组共计30个样本作为预测集。鉴于甘蓝叶面不平整、皱褶较多，叶片颜色深浅不一等因素会给近红外光谱分析带来干扰，给预测模型的建立增加难度，提出一种光谱全波段平均分组积分（求和）预处理方法，将光谱波段平均分成n组，再对分组后每组数据积分求和，用预处理后的数据训练BP神经网络。实验表明，光谱全波段平均分组积分（求和）预处理方法，对光谱反射率一阶导数（FD）且分组数为25的神经网络训练效果最好，建模集识别准确率为97.50%，预测集识别准确率为96.67%，建模效果优于通常采用的提取光谱敏感、特征波段建模方法（建模集识别准确率为91.67%）。光谱全波段平均分组积分预处理方法在保留光谱数据更多特征波段的同时探索更多潜在敏感波段，能够降低光谱数据维度，减小单个光谱数据噪声对建模效果的影响，选择合适的分组数n，能取得较好的建模预测效果。

Qualitative Analysis of Chlorpyrifos Pesticide Residues in Cabbage Leaves Based on Visible Near Infrared Spectroscopy

Chlorpyrifos is one of the most widely used organophosphorus pesticides(OPs)in agricultural production.However,pesticide residues caused by excessive OPs pose a serious threat to the natural environment and human life and health.Therefore,it is of great significance to develop a rapid,accurate,convenient,and economic method for directly detecting OPs residues in agricultural products.Four groups of chlorpyrifos pesticide solutions with different volume concentrations(1∶200,1∶500,1∶800,1∶1000)were prepared,the control group was treated with pure water.Cabbage leaves were soaked in chlorpyrifos pesticide solutionsfor 3 minutes,30 leaf samples were collected from each group,and 150 samples were collected from 5 groups.The spectrum information of Chlorpyrifos in cabbage leaves was obtained by visible near-infrared spectroscopy(NIR),and the qualitative analysis of chlorpyrifos pesticide residues in cabbage leaves was carried out.In modeling,24 samples in each group,120 samples of 5 groups are taken as modeling training set,6 samples in each group and 30 samples of 5 groups are taken as prediction set.The near-infrared spectrum analysis will be interfered with by factors such as uneven leaf surface,more wrinkles and different color of cabbage leaves,which makes the establishment of prediction model more difficult.In this paper,an all-band average grouping integration preprocessing method is proposed.The spectral bands are averagely divided into n groups,and then each group of data is integrated as new data for neural network modeling.The experimental results show that the all-band average grouping integration preprocessing method proposed in this paper has the best modeling effect using the spectral reflectance first derivative(FD)when the group number is n=25.The modeling set recognition accuracy is 97.50%,and the prediction set recognition accuracy is 96.67%.The modeling effect is better than the commonly used spectral sensitive and characteristic band modeling method(with modeling set recognition accuracy 91.67%).The all-band average grouping integration preprocessing method can retain more characteristic bands of spectral data and reduce the dimension of spectral data,reducing the impact of single spectral data noise on the modeling effect.Selecting the appropriate grouping number could achieve good modeling and prediction effect.The results of this study can provide a reference for the application of visible near-infrared spectroscopy in the detection of chlorpyrifos pesticide residues.