基于随机NMF理论的高光谱端元抽取  

Hyperspectral endmember extraction based on random NMF theory

作　　者：刘雪松 谭文群[1] 彭天亮 LIUXuesong;TAN Wenqun;PENG Tianliang(Jiangxi Province Key Lab of Water Information Cooperative Sensing and Intelligent Processing,Nanchang Institute of Technology,Nanchang 330099,China)

机构地区：[1]南昌工程学院江西省水信息协同感知与智能处理重点实验室,江西南昌330099

出　　处：《南昌工程学院学报》2020年第1期86-92,共7页Journal of Nanchang Institute of Technology

基　　金：国家自然科学基金资助项目(61701215);江西省教育厅自然科学基金资助项目(GJJ161132);江西省重点实验室开放基金项目(2016WICSIP027);南昌工程学院大学生科研训练计划(2018)。

摘　　要：高光谱端元抽取是光谱解混的一部分。非负矩阵分解(NMF)理论由于其能够很好的保持数据的非负特性,被广泛的应用于光谱解混。随着高光谱设备的提高,摄像机的光谱频带数也从几百个频带扩展到几千个频带。从而使光谱数据也越来越大。经典NMF理论是块处理法,数据越大,计算量也大幅增加。为了加快光谱解混速度,提出了基于随机L 1/2 NMF理论的端元抽取方法,与经典NMF相比,该方法能够快速提升算法速度,并且能够保证很好的解混精度。最后,用真实仿真数据验证了随机NMF理论的有效性。Hyperspectral endmember extraction is part of spectral unmixing.The non-negative matrix factorization(NMF)theory is widely used in spectral unmixing because it can well maintain the non-negative properties of data.As hyperspectral devices increase,the number of spectral bands in a camera also extends from hundreds of bands to thousands of bands.As a result,the spectral data is also growing.The classical NMF theory is a block processing method.The larger the data,the larger the calculation amount.In order to speed up the spectral unmixing speed,this paper proposes an endmember extraction method based on random L 1/2 NMF theory.Compared with the classical NMF,this method can quickly improve the algorithm speed and ensure good unmixing precision.Finally,we validate the validity of the random NMF theory with real simulation data.

关 键 词：端元抽取 光谱解混 非负矩阵分解 随机 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]