论文部分内容阅读
本研究目的在于分析农药残留量(pesticide residue,PR)与高光谱中响应特征参数之间的关系,并利用筛选的光谱特征参数建立反演毒死蜱残留量的有效模型。首先采用ASD Fieldspec高光谱仪测得韭菜样本的光谱,通过气相色谱-质谱联用(GC-MS)法测得毒死蜱残留量(PR)值;分析样本光谱反射率值及其一阶微分值与毒死蜱残留量的相关性,计算33个高光谱特征参数与毒死蜱残留量的相关性;根据相关系数高低选择敏感的光谱特征参数;最后采用最佳相关系数下的光谱特征参数对毒死蜱残留量进行建模反演。相关性分析结果显示:近红外波段789~867 nm范围内一阶微分光谱值与PR值呈正相关,1 860 nm处一阶微分光谱值(first-order differential 1 860 nm,FD1860)与PR值紧密相关;在33个高光谱特征参数中,近红外一阶微分总和(the sum of first-order differential near infrared,SDnir)与PR值呈良好的正相关关系。基于此,文章以供试样本的FD1860和SDnir观测值为自变量,分别建立了3个预测毒死蜱残留量的模型,即线性、二次多项式及指数模型,并采用交叉验证测试方法检验了模型的合理性。对实验所得决定系数R2和预测均方根误差(RMSE)的评价结果表明,以SDnir为自变量构建的模型稳定性强,其二次多项式模型是最佳反演毒死蜱残留量的有效模型。因此,样本的高光谱特征参数SDnir的变化幅度直接反映了韭菜样本中毒死蜱残留量的变化,表明运用蔬菜的高光谱特征参数反演蔬菜中农药残留量的方法是可行的。
The purpose of this study was to analyze the relationship between pesticide residues (PR) and response parameters in hyperspectral spectra and establish an effective model for the determination of residual chlorpyrifos residues by using the selected spectral parameters. The spectra of Chinese chive samples were measured by ASD Fieldspec hyperspectral spectrophotometer. The residual chlorpyrifos (PR) values were determined by gas chromatography-mass spectrometry (GC-MS). The spectral reflectance and first- The correlations between the parameters of 33 hyperspectral parameters and the residues of chlorpyrifos were calculated. Sensitive spectral parameters were selected according to the correlation coefficients. Finally, the chlorpyrifos residues were modeled using the spectral characteristic parameters of the best correlation coefficient Inversion. Correlation analysis showed that there was a positive correlation between the first-order differential spectral values in the near infrared band and the PR value in the range of 789-867 nm, the first-order differential 1 860 nm (FD 1860) at 1860 nm and the PR value Among the 33 hyperspectral parameters, the sum of first-order differential near infrared (SDnir) and PR values showed a good positive correlation. Based on this, three models of predicting chlorpyrifos residues, ie, linear, quadratic polynomial and exponential model, were established based on the observed values of FD1860 and SDnir of the samples tested, and the cross validation method was used to test the model rationality. The results of the experimental determination of the coefficient of determination R2 and the root mean square error of prediction (RMSE) show that the model built with SDnir as an independent variable is robust and the quadratic polynomial model is the most effective model for validating the residual chlorpyrifos residues. Therefore, the change range of SDnir, which is the hyperspectral parameter of sample, directly reflects the change of chlorpyrifos residues in Chinese chive samples. It is feasible to use the hyperspectral parameters of vegetables to retrieve the pesticide residues in vegetables.