Fast endmember extraction method using the geometry of the hyperspectral datacube

Proceedings of SPIE(2011)

引用 0|浏览2
暂无评分
摘要
This paper proposes a new method to extract the endmembers of a hyperspectral datacube using the geometry of the datacube. The criterion used to find the endmembers in this method is the volume of the simplex. Unlike to the widely used endmember extraction method "N-FINDR", which calculates the volume of a simplex as many times as the number of the vertices of the simplex for each pixel of the datacube in searching for the replacers for the vertices, the proposed method calculates the volume only once for each pixel of the datacube by taking into account of the geometry of the hyperspectral datacube that is tackled. For each pixel, the proposed method finds the closest vertex of the simplex to that pixel. Then the closest vertex is replaced with the pixel for updating the simplex. Computational complexity of the proposed method is one order of magnitude less than the N-FINDR. As the proposed method is using the same criterion as N-FINDR we refer it to as fast N-FINDR (FN-FINDR). The performance of the proposed method was compared with N-FINDR using an AVIRIS datacube and a HYDICE datacube. The performance of the proposed method was evaluated using three different distance measures. The comparison was also made using two different dimensionality reduction methods. It is observed that the FN-FINDR with a modified Euclidean distance works as well as N-FINDR.
更多
查看译文
关键词
Hyperspectral imagery,endmember extraction,spectral unmixing,N-FINDR
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要