地沟油的高光谱数据聚类分析  

Clustering analysis based on hyperspectral DN values of waste oil

作　　者：郭毅[1] 丁海勇[1] 徐晶鑫 徐灏[1] 

机构地区：[1]南京信息工程大学遥感学院,南京210044

出　　处：《国土资源遥感》2014年第1期37-41,共5页Remote Sensing for Land & Resources

基　　金：科研启动基金项目(编号:20090207);大学生创新训练基金项目(编号:N1885012178);国土环境与灾害监测国家测绘局重点实验室开放基金项目(编号:LEDM2010B06)共同资助

摘　　要：为了利用油品光谱特性的差异鉴别食用油和地沟油，以2种地沟油和4种食用油混合组成的22个样品为对象，研究了高光谱DN值和聚类分析法相结合鉴别食用油和地沟油的可行性。首先利用样品在350～2500nm范围内的光谱特征，选择高光谱DN值进行聚类分析；然后对DN值进行预处理，利用相关性距离法、欧氏距离法、标准化欧氏距离法及明可夫斯基距离法计算数据矩阵中对象之间的距离，用最短距离法、最长距离法、未加权平均距离法等8种距离公式计算系统聚类树；最后对各种方法生成的聚类树进行对比。研究结果表明，采用相关性最短距离法及内平方和距离法用于聚类分析，可准确地将油品分为22类，能够精确区分样本中的各类油品。In order to classify the edible oil and the waste cooking oil by using their difference in spectral characteristics, the authors employed 22 samples collected from the mixture of two kinds of waste oil and four sorts of edible oil to analyze the possibility of distinguishing these two kinds of oil by clustering their hyperspectral digital number. Spectral data which lies in the range of 350 - 2 500 nm were utilized in this paper for clustering analysis. Digital number of hyperspectral data, first order derivation and second order derivation of the reflective data were used as the spectral information for the target. Correlation distance, Euclidean distance, standardized Euclidean distance and Minkowski distance method were used to calculate the distance between the spectral objects in the data matrix. And then, eight different kinds of distance method were employed to compute the clustering tree, which accurately classified these oils into twenty- two types. Numerical experiments demonstrate that un- weighted distance method and interior square sum distance could be utilized in the correlation clustering analysis to accurately distinguish different kinds of oils in the sample and to classify these oils into 22 types accurately.

关 键 词：地沟油 高光谱 最短距离法 聚类分析 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]