以凸面单体边界为搜索空间的端元快速提取算法

Fast extraction of endmembers from convex simplex's boundary

提出了以凸面单体边界为搜索空间的端元快速提取算法,其核心包括凸面单体边界的确定和以凸面单体边界为基础的端元搜索两部分。实验表明:该算法不仅能够准确地寻找到端元,而且端元提取速度明显快于现有的端元提取算法。

Endmember extraction is one of the key problems for mixel classification of multispectral imagery.Existing algorithms based on convex simplex often find endmembers within the whole convex simplex so that their speeds are slower when more samples are used to obtain endmembers.Since only the vertexes of convex simplex are probably endmembers and they must be located in the boundary of convex simplex,the search space will shrink a lot if finding endmembers is performed only within the boundary points of convex simplex.According to this theory,this paper presents the endmember extraction algorithm based on the boundary of convex simplex.The algorithm includes determination of boundary of convex simplex and fast finding endmembers within the boundary points of convex simplex.Experiments show that the algorithm can find endmembers not only correctly but also faster than existing algorithms.