中国科学院遥感与数字地球研究所-高光谱遥感应用技术研究室

【作者】             白继伟                                                                                                                                                  

【导师】             童庆禧；郑兰芬

【 学位年度 】    2002

【论文级别】      硕士    

【关键词】          高光谱   数据库   光谱匹配   包络线消除

【Key words】   hyperspectral database    spectral mapping    continuum removed

【中文摘要】                                                                                                                                                             

 高光谱分辨率遥感具有图像——光谱合一的特点，是对传统遥感数据源如航片，多光谱遥感数据等的有力补充。高光谱分辨率遥感数据以其丰富的光谱信息使得其分析处理集中于光谱维上进行图像信息的展开和定量分析。在高光谱图像处理中，光谱匹配分类技术是高光谱地物识别的关键技术之一。本文的研究目的是针对高光谱的特点建立实用、有效的数据库，同时在此基础上，对比分析几种常用的光谱匹配分类技术的优缺点，并验证一种新型的匹配分类技术。文章首先简要介绍了现有的高光谱数据库及其重要作用；然后利用现有高光谱数据库中的标准数据，对现有的一些典型的基于光谱数据库的光谱匹配技术进行研究，对比分析其识别效果，并就分类的结果进行探讨；最后，注意到在高光谱遥感中，包络线消除法作为一种有效的光谱分析工具，但一般仅局限于对单个像元的光谱进行光谱分析，从中提取出有助于分类识别的特征波段的现象，本文发展并验证了一种新型的基于高光谱数据库的光谱匹配分类方法：以包络线消除算法为基础，应用VC++语言编程实现了对整个高光谱图像文件去包络、归一化并且提取出分类的特征空间的功能,并且针对原图像文件和去包络线后的图像文件，比较了应用最大似然分类法和光谱角度匹配法进行分类的结果。试验表明，新型方法对于光谱曲线具有明显吸收和反射特征的地物识别是比较有效的，但对于光谱曲线平缓，没有明显特征的地物则难于使用，有待于对其进行进一步的改进。 

【Abstract】

The data of the hyperspectral remote sensing has the characteristic of combining the image and spectrum, which is the powerful supplement for the traditional data resource such as aerial picture and multi-spectral image. We study on the image information and quantitative analysis focusing on the spectral dimension because the hyperspectral data has the abundant spectral information. During transacting the hyperspectral image, the spectral mapping analysis is one of the key techniques to identify and classify the material in the image. In this paper, I aim to build a useful and effective database fitting with the hyperspectral data, then, on the base of this database, contrast and value some spectral mapping technique in common use, and then develop and validate a new spectral mapping technique. So at first this paper introduces the existing hyperspectral database briefly and its important functions, then with the standard spectrums in the existing database, do some study on the value of some typical spectral mapping technique and discuss the results of the different techniques, at last, noticed that in the hyperspectral remote sensing, the continuum removed method is very useful but used only with the spectrum of a single pixel to analyze spectrum and extract the feature bands useful with the classification, so in this paper, based on the continuum removed arithmetic, I programmed with VC++ to fulfil the functions of the continuum removed to the whole hyperspectral image, normalization and extracting the feature space for the classification. Then, aiming at the former image and the after-continuum removed image, the classification results of the MLC and SAM are compared. The results showed that the new technique is effective to identify the material with the spectrum which has the distinct absorbing and reflecting characteristic, but it’s hard to use with the material having no characteristic of the spectrum. So this method needs to be improved in the future.