基于局部窗口的端元提取光谱优化方法
2016-11-02秦学川
秦学川
摘要:目前,已经存在的大多数端元提取方法都只利用了图像的光谱特性,而忽略了图像的空间信息。HEEA方法是少数利用图像空间信息进行端元提取的方法之一。而,HEEA方法在考虑图像空间信息时存在两个问题,一是直接利用像元坐标之间的欧氏距离计算光谱邻域权重,这样可能导致的后果是会为相差甚远的像元设置相同的光谱权重。二是该方法需要通过计算像元光谱之间的SID-SAD值,并将该值与事先设置好的阈值进行比较,从而确定一个像元是否被用来进行光谱优化。而设置不同的阈值会获得不同的结果。针对上述问题,本文提出了一种利用光谱角距离和欧氏距离计算光谱邻域权重的光谱优化方法。
关键词:端元提取;空间信息;光谱优化
中图分类号:TP311 文献标识码:A 文章编号:1009-3044(2016)18-0163-04
Spectral Optimizing Algorithm for Endmember Extraction Based On a Local Window
QIN Xue-chuan
(College of Information Science and Engineering, Shandong University of Science and Technology, Qingdao, 266590, China)
Abstract : At present, most of the existed endmember extraction methods are only using the spe-ctral characteristic of the image, while ignoring the spatial information of the image. HEEA meth- is one of the few endmember extraction method that using the spatial information of the ima-ge. But, there are two problems with HEEA method , one is directly using the Euclidean distance between pixels' coordinates to calculate spatial neighborhood weight, and that could lead to the consequence of setting the same spatial neighborhood weight for different spectrum. The second is that need to calculate SID-SAD value between different spectrum and to compare with a thres-hold so that to determine whether a pixel is used to carry out spectrum optimization. But ,differ-ent threshold can lead to different unmixing result. Aiming at these problems, a novel spectral optimizing algorithm that use spectral angle distance and Euclidean distance to calculate spatial-neighborhood weight is proposed in this paper.Key words: endmember extraction ; spatial information ;spectral optimizing
近年来,高光谱遥感影像的应用范围越来越广。但受限于高光谱影像的空间分辨率以及影像中地物分布的多样性,混合像元广泛存在于遥感影像中 [1]。混合像元是指包含多种地物的像素点。混合像元的存在给高光谱影像的进一步应用带来了诸多困难,为了解决混合像元的问题,端元提取技术应运而生。
在过去的几十年的时间里,大量的端元提取算法相继被提出。根据算法有没有考虑图像空间信息进行分类,可以分为两大类。一类为没有考虑空间信息的算法,这些算法包括纯像元指数算法(pure pixel index ,PPI) [2]、N-FINDR算法 [3]、顶点成分分析(vertex component analysis,VCA) [4]等。另一类为考虑了空间……
