基于多光谱融合的奶粉掺假诊断方法  

Adulterated Milk Powder Diagnosis Method Based on Multi-Spectra Fusion

作　　者：陈达[1] 骆文欣 黄志轩 李奇峰[1] 

机构地区：[1]天津大学精密仪器与光电子工程学院,天津300072

出　　处：《纳米技术与精密工程》2017年第5期384-388,共5页Nanotechnology and Precision Engineering

摘　　要：发展了一种多光谱融合新技术,该技术充分利用拉曼光谱与红外光谱的互补特性,并借助数据融合手段,高效实现奶粉掺假检测.为进一步提升数据融合算法的准确性,有机结合离散小波变换(DWT)多尺度特性及竞争性自适应重加权偏最小二乘线性判别(CARS-PLSDA)算法,以有效扣除光谱建模中的干扰信息.为验证多光谱融合技术的有效性,对4种典型奶粉掺假体系分别建立分类判别模型.结果表明,基于DWT-CARS-PLSDA多光谱融合算法所建的面粉、淀粉、糊精和大豆分离蛋白奶粉掺假模型灵敏度分别为94.74%、100%、84.21%和100%,正确率分别为99.42%、98.83%、98.25%和98.83%.与单独对拉曼光谱或红外光谱建立模型相比,4种模型能够显著提高奶粉掺假检测灵敏度和准确性,为奶粉掺假快速诊断提供了一种有效工具.To detect the adulteration of milk power more effectively and more rapidly, a multi-spectra fu-sion technique （ MFT） consisting of Raman and infrared spectrometries is developed to utilize their com-plementary molecular characteristics localized in their spectra. In MFT, a data fusion strategy is proposed to combine the discrete wavelet transform （ DWT） and competitive adaptive reweighted sampling-partial least squares discriminant analysis （CARS-PLSDA） , encoding the multi-spectra information in the pres-ence of complex spectral interference effectively. The DWT-CARS-PLSDA method is validated and re-fined by performing MFT analysis of 4 typical milk powder adulterant systems. For each model the satis-factory results are achieved with sensitivity of 94. 74% , 100% , 84. 21% and 100% respectively and ac-curacy of 99. 42% , 98. 83% , 98. 25% and 98. 83% respectively. The results indicate that the MFT out- performed solitary Raman spectroscopy or infrared spectroscopy in both detection sensitivity and accuracy, revealing that the MFT is a promising tool for discriminating the milk powder adulteration.

关 键 词：拉曼光谱 红外光谱 奶粉掺假 数据融合 DWT-CARS-PLSDA 

分 类 号：O657.3[理学—分析化学] TS252.51[理学—化学]