摘要
提出了基于CARS、SPA和CARS-SPA特征波长提取的激光近红外光谱技术快速鉴别食用植物油种类的方法。应用光谱仪采集127个食用植物油样本的光谱数据,采用标准正态变量变换算法(SNV)、标准正态变量变换和去趋势技术联用算法(SNV-DT)对其进行预处理,采用CARS、SPA和CARS-SPA 3种方法对经过预处理的光谱数据进行特征波长提取,应用支持向量机分类方法(SVC)建立食用植物油种类定性分类校正模型,选择网格搜索算法对模型参数组合(C,g)进行寻优,确定最优参数组合。结果表明,CARS-SVC、SPA-SVC和CARS-SPA-SVC模型预测集准确率均达到96.77%,预测效果理想,其中SNV-DT-SPA-SVC模型预测效果最优,预测集准确率达到100%。综上基于特征波长提取的激光近红外光谱分析技术能够快速准确鉴别食用植物油种类,为便携式现场检测设备开发提供了理论基础。
To classify edible vegetable oil quickly, the method of laser near- infrared（NIR） spectroscopy combined with extration of characteristic wavelength by heavy competitive adaptive weighted sampling （CARS）, successive projections algorithm（SPA） and CARS -SPA was proposed. Spectral data of 127 edible vegetable oil samples were collected by laser NIR spectrometer, and pretreated by standard normal variate transformation （ SNV）, standard normal variate transformation and de - trending （ SNV - DT）, then the characteristic wavelength was extracted using CARS, SPA and CARS -SPA, and the support vector machine classification （SVC） method was applied to establish qualitative classification correction model of edible vegetable oil. In the end, the model parameter combination （C, g） were optimized by mesh search algorithm. The results showed that the accuracy rates of prediction set of CARS - SVC, SPA - SVC and CARS - SPA - SVC models all reached 96.77% and the prediction effect was satisfying, in which the prediction effect of SNV - DT - SPA - SVC model was the best with the accuracy rate of prediction set 100%. NIR spectroscopy combined with extraction of characteristic wavelength could identify edible vegetable oil accurately and quickly, and they could provide theoretical basis for develop-ment of portable field testing equipment.
出处
《中国油脂》
CAS
CSCD
北大核心
2017年第4期72-75,共4页
China Oils and Fats
基金
国家"十一五"科技支撑计划项目(2009BADB9B08)
武汉市科技攻关计划项目(2013010501010147)
武汉工业学院食品营养与安全重大项目培育专项(2011Z06)
关键词
激光近红外光谱技术
食用植物油
特征波长提取
支持向量机分类
掺伪
laser near - infrared spectroscopy
edible vegetable oil
characteristic wavelength ex- traction
support vector machine classification
adulteration
