基于高光谱图像技术和波长选择方法的羊肉掺假检测方法研究  

Hyperspectral Imaging Technology and Wavelength Selection Method for Nondestructive Detection of Mutton Adulteration

作　　者：刘友华[1] 白亚斌[1] 邱祝福 陈伟[1] 冯耀泽[1] 

机构地区：[1]华中农业大学工学院,湖北武汉430070

出　　处：《海南师范大学学报（自然科学版）》2015年第3期265-269,共5页Journal of Hainan Normal University（Natural Science）

基　　金：湖北省自然科学基金重点项目(ZRZ2015000186);中央高校基本科研业务费专项基金资助项目(2662015PY051);华中农业大学新进博士科研启动基金

摘　　要：针对肉类掺假的定量检测问题,建立了基于高光谱图像技术结合波长选择方法以及偏最小二乘（PLS）法的羊肉掺假无损检测方法.试验搭建了羊肉-猪肉的高光谱图像采集系统,对选取的99个样本进行高光谱图像（390-1040nm）采集并提取感兴趣区域光谱.比较了多种预处理方法对全波长模型的影响,标准正态变量校正（SNV）为最优光谱预处理方法,校正决定系数R2CV达到0.93,预测决定系数R2P达到0.96,校正均方根误差RMSECV为8.33%,预测均方根误差RMSEP为6.18%.采用多种波长选择方法对光谱全波段520个变量进行变量选择,最终确定基于竞争性自适应重加权算法（CARS）的简化模型最优,其R2C=0.96,R2CV=0.94,R2P=0.96,RMSEC=6.55%,RMSECV=7.66%,RMSEP=6.10%.高光谱图像技术结合CARS能够对掺假羊肉进行准确的定量检测,可为羊肉掺假的在线无损检测提供理论依据。For the quantitative detection of mutton adulteration, hyperspectral imaging was combined with wavelength selection methods as well as partial least squares（PLS）method to establish calibration models. The test has set up a hyperspectral image acquisition system, and the hyperspectral images of 99 selected mutton samples were collected（390-1040nm）. Sample spectra were extracted from the regions of interest of samples. Comparison of the effects of various pretreatment methods for full wavelengths model was conducted and the test showed that the standard normal variable correction（SNV）was the optimal spectral pretreatment method with coefficient of determination R2 CVand R2Pof 0.93 and 0.96, root mean square error RMSECVand RMSEPof 8.33% and 6.18%, respectively. A variety of wavelength selection methods were utilized to select feature wavelengths and competitive adaptive re-weighted algorithm（CARS）led to the best simplified model where R2C=0.96,R2CV=0.94,R2P=0.96,RMSEC=6.55%,RMSECV=7.66%,RMSEP=6.10%. The results showed that hyperspectral imaging technology combined with CARS was powerful for accurate quantitative detection of mutton adulteration,and also can provide a theoretical basis for the on-line non-destructive testing of mutton adulteration.

关 键 词：羊肉掺假 标准正态变量校正 竞争性自适应重加权算法 偏最小二乘回归 定量分析 

分 类 号：O657.3[理学—分析化学]