论文部分内容阅读
高光谱图像中混合像元的存在不仅影响了基于遥感影像的地物识别和分类精度,而且已经成为遥感科学向定量化方向发展的主要障碍。本文分析和研究了现有的典型端元提取算法,在此基础上,对这些算法进行归纳总结,从是否假定纯像元存在角度将其分为两类:端元识别算法和端元生成算法,并就两种分类方法选取了具有代表性的6种典型端元提取算法:N-FINDR、VCA、SGA、OSP、ICE和MVC-NMF算法进行分析和实验。通过对这6种方法的实验比较,得出两种端元提取分类方法的优点与不足,并对今后的研究工作提出展望。
The existence of mixed pixels in hyperspectral images not only affects the accuracy of object recognition and classification based on remote sensing images, but also has become the major obstacle to the quantitative development of remote sensing science. This paper analyzes and studies the existing typical endmember extraction algorithms. Based on these, we summarize these algorithms and divide them into two categories from the point of view of existence of pure pixels: endmember recognition algorithm and endmember generation algorithm , And six typical representative endmember extraction algorithms are selected for the two classification methods: N-FINDR, VCA, SGA, OSP, ICE and MVC-NMF algorithms for analysis and experiment. Through the comparison of the six methods, the advantages and disadvantages of two kinds of endmember extraction and classification methods are obtained, and the future research work is put forward.