一种非监督快速端元提取方法  

An unsupervised and fast method of endmembers extraction

作　　者：陈伟[1] 余旭初[1] 刘伟[1] 杨国鹏[1] 

机构地区：[1]信息工程大学测绘学院,郑州450052

出　　处：《测绘科学》2009年第5期147-149,共3页Science of Surveying and Mapping

摘　　要：正交子空间投影(OSP)方法广泛用于目标与背景的分离之中,对于高光谱影像,OSP可用于目标提取和混合像元分解,但缺点是需要端元的先验知识。针对这一问题,本文基于OSP的原理提出了一种非监督快速端元提取方法。实验使用模拟高光谱数据和由OM ISⅠ获取的真实高光谱数据,结果精度令人满意,证明了本文算法进行端元自动提取的可行性。Orthogonal subspace projection （OSP） is widely employed to separate the targets and background, extract targets and mixed pixels unmixing from hyperspectral imageries. However, there is a disadvantage that OSP needs full prior knowledge of endmembers. This article brings forward an unsupervised and fast method to extract endmembers from hyperspectral imageries according to the principle of OSP. The paper carries out the endmember extraction experiment by simulative hyperspectral data and real hyperspectral data collected by OMIS I. The result of experiment is satisfying and indicates that the algorithm is feasible to automatically endmembers extraction from hyperspectral imageries.

关 键 词：高光谱影像 正交子空间投影 最小二乘 像元分解 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]