中文摘要：

通过分析线性混合模型及几何端元在空间投影中所具有的特性，提出了用于高光谱图像端元提取的分层查找法。该方法基于几何形态学与几何端元的概念，将确定单形体端点的过程分层处理，以实现对高光谱图像中端元的快速、准确估计。该方法不需要预先确定端元个数，而是在提取端元过程中自动调整端元个数。基于模拟数据与真实数据的实验的结果表明，在端元个数未知的情况下，分层查找法能够对高光谱图像中所包含的端元及端元个数给出较准确的快速估计，较好地解决了实际高光谱图像端元提取过程中端元个数难以确定的问题。

英文摘要：

The paper presents a hierarchical approach to extraction of endmembers in hyperspectral images through close analyses of the linear mixed model for mixed pixels and the properties of geometric endmembers in space projection. For fast and correct estimation of endmembers in hyperspectral images, the approach separates the process of finding the extreme points of simplex into several steps based on the concepts of geometric morphology and geometric endmember. Besides, it does not need to know the number of endmembers, but modifies the number of endmembers automatically during the process of endmember extraction. Some results of the experiments using both the simulated hyperspectral data and the real hyperspectral data show that the hierarchical approach can give good estimates of the endmembers when the number of endmembers is unknown, so can successfully handle the problem that the number of endmembers in real hyperspectral data is often difficult to determine.