欧亚国际

摘要

背景

牛奶和牛奶产品中的双氰胺(DCD）污染日益成为紧迫以及广受关注的议题，尤其在发了多起严重食品有效事故后。该研究使用了多光谱成像法(405–970 nm) （Videometerlab多光谱成像仪）联合化学计量法检测婴儿配方奶粉中的双氰胺DCD的方法。偏较小二乘法（PLS），较小二乘支持向量机(LS-SVM)以及BP神经网络(BPNN)用来开发定量模型。

结果

与PLS和 LS-SVM法相比，BPNN显著的提升了预测性能，预测测定系数分别= 0.935 和 0.873，1段和2段婴儿配方奶粉剩余预测偏差分别为(RPD) = 3.777和3.060 。另外，多光谱成像可区分纯婴儿配方奶粉和掺有0.01% DCD 双氰胺的奶粉，使用BPNN模型，区分无误。

结论

研究显示，多光谱成像结合化学计量法可实现在婴儿配方奶粉中的DCD双氰胺快速无损检测。

Non-destructive detection of dicyandiamide in infant formula powder using multi-spectral imaging coupled with chemometrics

Abstract

BACKGROUND

Dicyandiamide (DCD) contamination of milk and milk products has become an urgent and broadly recognised topic as a result of several food safety scares. This study investigated the potential of using multi-spectral imaging (405–970 nm) coupled with chemometrics for detection of DCD in infant formula powder. Partial least squares (PLS), least squares-support vector machines (LS-SVM), and back-propagation neural network (BPNN) were applied to develop quantitative models.

RESULTS

Compared with PLS and LS-SVM, BPNN considerably improved the prediction performance with coefficient of determination in prediction () = 0.935 and 0.873, residual predictive deviation (RPD) = 3.777 and 3.060 for brand 1 and brand 2 of infant formula powders, respectively. Besides, multi-spectral imaging was able to differentiate unadulterated infant formula powder from samples containing 0.01% DCD with no misclassification using BPNN model.

CONCLUSION

The study demonstrated that multi-spectral imaging combined with chemometrics enables rapid and non-destructive detection of DCD in infant formula powder.