应用近红外光谱分析技术检测豆粕中掺杂三聚氰胺和尿素的研究

The application of near-infrared reflectance spectroscopy(NIRS) to detect melamine and urea adulteration of soya bean meal

利用近红外光谱分析技术对豆粕中掺杂的三聚氰胺和尿素进行判别鉴定及三种模式识别方法的比较。本研究收集了30批次的纯豆粕,共180个样品,分别在纯豆粕中添加三聚氰胺和尿素,制备掺假物含量不同的豆粕样本共214个,掺假物的质量浓度范围为0.1%～5%。分别用有监督模式方法簇类独立软模式（SIMCA）、偏最小二乘判别分析（PLS-DA）和支持向量机（SVM）建立豆粕的判别模型。结果显示,SVM方法的校正集准确率为100%,验证集准确率为98.1%,是最有效的分类方法;其次是PLS-DA方法,掺入三聚氰胺的模型的校正集准确率为96.9%,验证集准确率为95.3%,掺入尿素的模型的校正集准确率为94.5%,验证集准确率为94.2%;而SIMCA各模型的识别率高,但拒绝率较低,可仅对纯豆粕进行主成分分析,其识别率达到92.2%,拒绝率为88.5%。因此,近红外光谱法能用于掺杂三聚氰胺和尿素的豆粕的快速筛选。

In order to study the feasibility of using near infrared reflectance spectroscopy （NIRS） to discriminate adulteration of soya bean meal and the comparison of three pattern recognition methods, a total of 180 samples were collected from 30 batches of pure soya bean meal. We prepared 214 specimens that were adulterated with different proportion （0.1%-5%） of melamine and urea respectively. Discriminate models were built by different supervised pattern recognition techniques： soft independent modeling of class analogy （SIMCA）, partial least squares diseriminant analysis （PLS-DA） and sup- port vector machine （SVM）. The results obtained show that SVM is the most effective techniques with 100% classification accuracy in calibration set and 98.1% in validation set followed by PLS-DA with the accuracy of 96.9% and 94.6% for discriminating melamine adulteration and 94.5% and 94.2% for discriminating melamine adulteration while SIMCA model showed relative high recognition rate and low rejection rate. But when the pure soya bean meal set was just applied principal component analysis, recognition rate and low rejection rate was 92.2% and 88.5%, respectively. Therefore, NIRS was a rapid discriminating method for screening adulteration of soya bean meal with melamine and urea.