利用光谱趋势参数快速判定小麦粉DON等级的研究

The Study on Quickly Determining DON Level in Wheat Flour by Trend Parameter of Spectra

脱氧雪腐镰刀菌烯醇(DON)是一种经常发生在谷物及其衍生产品中的霉菌毒素,危害人类和家畜的生命健康。开发快速、准确、经济、无环境危害的检测方法是一个亟待解决的问题。本研究定义了可见光-近红外(Vis-NIR)光谱的趋势参数TP(trend parameter),利用TP确定与DON浓度最相关的特征波段。文中校正集样本的光谱矩阵行按样本DON浓度逐渐增加的顺序排列,矩阵每一列(每一个波段)都对应一个TP值,所有样本在某波段下的吸光度在列方向上的递增趋势越强(即TP值越大),则此波段下的吸光度与DON浓度的相关性就越强,该波段便可以作为评估DON浓度的特征波段。研究发现在666, 1 238和1 660 nm处TP出现局部最大值,利用此三个特征波段下的光谱进行二次判别分析Quadratic Discriminant Analysis(QDA),以此构建的TP-QDA模型可以将小麦粉按DON污染水平分成轻度(0<DON<1 000μg·kg-1)、中度(1 000≤DON<2 000μg·kg-1)、和重度(DON≥2 000μg·kg-1)三个等级。该模型的整体分类准确率在校正集和验证集中分别为88.24%和86.27%。对比了传统的主成分分析PCA(principal component analysis)特征波段选取方法,其所构建的PCA-QDA模型也将相同的小麦样品分成三个污染等级,整体分类准确率在校正集和验证集中分别为68.62%和72.55%。这些研究结果证实了TP选择特征波段的方法在判断DON污染水平时优于PCA特征波段选取方法,并且TP-QDA模型可有效地用于快速对小麦粉的污染等级进行分类,从而减少收储运过程中分析筛选小麦的时间和经济成本。研究结果还有待在更广泛的小麦品种中进行普适性验证。

Deoxynivalenol(DON) is a mycotoxin that often occurs in cereals and their derivatives. It is harmful to the life and health of human and livestock. It is urgent to develop a detection method, which can rapidly, accurately and economically detect DON without environmental hazard. This study defined a Trend Parameter(TP) of the visible-near-infrared(VIS-NIR) spectra. The TP was used to determine the characteristic bands which were most relevant to the DON concentration. In this paper, the rows of spectral matrix of the samples in calibration set were arranged in the order of gradual increase in DON concentration. Each column(each band) of the matrix corresponded to a TP value. Under a certain band, the stronger the increasing trend of the absorbances of all samples in the column direction is(ie, the larger the TP value), the stronger the correlation between the absorbance and the DON concentration in this band is, and this band can be used as a characteristic band for evaluating the DON concentration. The study found that the local maximum of TP appeared at 666, 1 238, and 1 660 nm. The quadratic discriminant analysis(QDA) was performed by the spectra of the three characteristic bands. The wheat flour can be divided into three grades: mild(0<DON<1 000 μg·kg-1), moderate(1 000≤DON<2 000 μg·kg-1), and severe(DON≥2 000 μg·kg-1) pollution by the constructed TP-QDA model. The overall classification accuracy of the model was respectively 88.24% and 86.27% in the calibration set and verification set. The Principal Component Analysis(PCA) of characteristic bands selection was used to make a comparison. The PCA-QDA model divided the same wheat sample into three pollution levels. The overall classification accuracy rate was 68.62% in the calibration set and 72.55% in the verification one. These findings confirmed that the selection of characteristic bands by TP parameter is superior to the one by the PCA in judging DON pollution level, and the TP-QDA model can be effectively used to quickly classify the pollution level of wheat flour, thereby reducing the time and economic cost of analyzing and screening wheat during the process of acquisition, storage and transport. The results of this study have yet to be tested for universality in a wider range of wheat varieties.