摘要
光谱相似性测度是高光谱影像光谱匹配分类的准则,但单一的光谱相似性测度不能综合考虑光谱曲线的形状、辐射等多种特征,因此将其用于高光谱影像光谱匹配分类时精度较低,研究发现两种或多种相似性测度的组合能够有效提高光谱匹配分类的精度。在光谱角余弦测度和相关系数测度的基础上,采用算术平均的组合方式,得到光谱角余弦-相关系数测度,将其用于光谱匹配分类,实现光谱角余弦与相关系数测度组合的光谱匹配分类方法。提出了基于光谱角余弦-相关系数测度的光谱匹配分类流程,通过ROSIS和OMIS两组高光谱影像分类实验表明,相比于光谱角余弦测度和相关系数测度,将光谱角余弦-相关系数测度用于光谱匹配分类能够得到较高的总体分类精度,对单一地物的分类精度也有一定程度的改善。
Spectral similarity measure is the criterion for spectral matching classification of hyperspectral imagery.Single spectral similarity measure can not involve multiple features of spectral curves,such as shape feature and radiation feature,so classification accuracy is low when single spectral similarity measure is used to spectral matching classification of hyperspectral imagery.The study found a combination of two or more similarity measures could effectively improve the accuracy of spectral matching classification.On the basis of the measures of spectral angle cosine and spectral correlation coefficient,spectral angle cosine-spectral correlation coefficient measure was obtained by way of the arithmetic mean of the combination.By using spectral angle cosine-spectral correlation coefficient measure to spectral matching classification,spectral matching classification approach combined with spectral angle cosine and spectral correlation coefficient was achieved.According to the approach,spectral matching classification process based on spectral angle cosine-spectral correlation coefficient measure was proposed.Classification of ROSIS and OMIS hyperspectral imagery in two experiments showed that,compared with the two separated measures,a higher overall classification accuracy could be obtained when spectral angle cosine-spectral correlation coefficient measure was used to spectral matching classification,and there was also an improvement in single-object classification to some extent.
出处
《地理与地理信息科学》
CSCD
北大核心
2016年第3期29-33,2,共5页
Geography and Geo-Information Science
基金
地理信息工程国家重点实验室开放研究基金项目(SKLGIE2015-M-3-1
SKLGIE2015-M-3-2)
关键词
高光谱影像
光谱角余弦-相关系数测度
分类
精度
hyperspectral imagery
spectral angle cosine-spectral correlation coefficient measure
classification
accuracy
