SVM法定性鉴别奶牛饲料中脲醛树脂的可行性研究

Feasibility study of qualitative identification urea-formaldehyde resins in milk cow feed by support vector machine

收集中国常用的、具有代表性的奶牛精补料44个样品,制备176个脲醛树脂(urea-formaldehyde resins,UF)掺假样品。在全光谱范围内进行近红外透反射光谱扫描,选择不同的归一化方式进行前处理,采用支持向量机(supportvector machine,SVM)方法,筛选最佳的预处理方法来建立定性鉴别模型。当采用归一化方式与主成分分析(principal component analysis,PCA)相结合时,所建立的SVM定性分析模型的预测精确率达到97.701 1%。说明利用近红外透反射光谱建立定性分析模型来检测奶牛饲料中是否掺有UF的研究是可行的。

Forty four samples of representative milk cow concentrate supplement were collected, and 176 UF adulterated samples were prepared. The near-infrared transmission and reflectance spectra were scanned in the full spectrum. The qualitative identification mode was established by considering the different normalization ways, using SVM （Support Vector Machine）and getting the optimum pretreatment. When normalization ways combined with PCA （Principal Component Analysis）, the accuracy of prediction of SVM qualitative identification mode got to 97. 701 1 %. Experiments show that it is feasible to distinguish the milk cow concentrate supplement adulterated with UF from the pure samples by means of the qualitative analysis model established by near-infrared transmission and reflectance spectroscopy.