论文部分内容阅读
利用近红外光谱技术结合变量选择方法对食用油中高效氟吡甲禾灵残留进行定性检测研究.在4000～10000 cm-1光谱范围内采集114个食用油样本的近红外透射光谱.分别采用竞争自适应重加权法(Competitive Adaptive Reweighted Sampling,CARS)、子窗口重排分析(Subwindow Permutation Analysis,SPA)和蒙特卡罗无信息变量消除(Monte Carlo Uninformation Variable Elimination,MC-UVE)3种变量选择方法在全波段范围内筛选出与食用油中高效氟吡甲禾灵相关的重要变量,最后应用偏最小二乘-线性判别(Partial Least Squares-Linear Discriminant Analysis,PLS-LDA)方法分别对筛选后的特征波数变量建立食用油中高效氟吡甲禾灵残留的判别模型,并与常用定性判别方法的结果进行比较.研究结果表明,近红外光谱技术结合变量选择方法定性检测食用油中高效氟吡甲禾灵残留是可行的,且检测精度高.CARS-PLS-LDA方法所建立的判别模型性能最优,其预测集的正确率、灵敏度及特异性均为100.00％,且建模所用波数变量数最少,仅为全波段的0.82％.此外,CARS方法优于SPA及MC-UVE方法,但3种方法均能有效筛选关键变量,减少建模波数变量数,简化判别模型,提高判别模型的精度及稳定性.“,”In this study , near infrared spectroscopy combined with variable selection method is used for the qualitative detection of haloxyfop-p-methyl residue in edible oil .Near infrared transmission spectra of 114 edible oil samples are collected in the spectrum range of 4000～10000 cm-1 .The important variables related to haloxyfop-p-methyl residue in edible oil are selected from the whole band by competitive adaptive reweighted sampling ( CARS ) , subwindow permutation analysis ( SPA ) and monte carlo uninformation variable elimination (MC-UVE), respectively.Finally, partial least squares linear discriminant ( PLS-LDA ) method is applied to establish the discriminant model which uses selected characteristic wave variables of haloxyfop-p-methyl residue in edible oil .The results are compared with those of commonly used qualitative methods .The results show that the near infrared spectroscopy combined with variable selection method is feasible to detect haloxyfop -p-methyl residue in edible oil , and the accuracy of detection is high .The discriminant model established by CARS-PLS-LDA method exhibits best performance , with the correct rate , sensitivity and specificity of the prediction at 100.00％.The number of variables used in modeling is the least , and the number is only 0.82％of the full band.In addition, the CARS method is better than SPA and MC-UVE method.But all the three methods can not only select the key variables and reduce the number of wavelength variables used in model , but also simplify the discriminant model and improve the accuracy and stability of the discriminant model .