基于超像素分割和纯像元指数的端元束提取算法

An Endmember Bundle Extraction Algorithm Based on Superpixel Segmentation and Pure Pixel Index

摘要: 为了在高光谱图像中提取空间分布合理且冗余度小的可变端元，本文提出基于超像素分割和纯像元指数的端元束提取算法。首先对高光谱图像进行主成分变换，得到三个主分量并对图像进行基于熵率的超像素分割。通过纯像元指数法提取初始候选端元，每个超像素内只保留均质性指数最小的端元，然后通过聚类得到端元束，最后去除同类端元束内的冗余端元。仿真和真实数据结果表明，对比已有的端元束提取算法，本文提出的方法能更有效提取可变端元和减少端元冗余度。

Abstract: In order to extract variable endmembers with reasonable spatial distribution and small redun-dancy, an endmember bundle extraction algorithm for hyperspectral image based on superpixel segmentation and pure pixel index is proposed. First, the principal component transform is con-ducted on the hyperspectral image, and three principal components are obtained. The image is di-vided by the entropy-based superpixel segmentation. By preforming pure pixel index to extract initial candidate endmembers, the endmembers with the smallest homogeneity indices in each superpixel are retained, and then the endmember bundles are obtained by clustering. Finally, the redundant endmembers in the same bundle are removed. The simulated and real data results show that compared with the existing endmember bundle extraction algorithms, the proposed method can extract variable endmembers more effectively and reduce endmember redundancy.