高光谱遥感影像端元提取算法研究进展及分类  

Research Progress on Endmember Extraction Algorithm and Its Classification of Hyperspectral Remote Sensing Imagery

作　　者：王茂芝[1,2] 徐文皙[1] 王璐[1] 郭科[1] 

机构地区：[1]成都理工大学数学地质四川省重点实验室,四川成都610059 [2]电子科技大学航空航天学院,四川成都611731

出　　处：《遥感技术与应用》2015年第4期616-625,共10页Remote Sensing Technology and Application

基　　金：中国地质调查局地调项目(1212011120226);四川省教育厅自然科学重点项目"基于集群和GPU的高光谱遥感影像并行处理"(13ZA0065)

摘　　要：在给出端元的物理、代数和几何学解释基础上,对现有端元提取算法从算法设计机理出发,分为基于几何学、基于统计学和信号检测理论以及空间和光谱相结合三大类,并进一步对基于几何学的端元提取算法从技术处理手段差异细分为基于距离、体积、投影变换和最优化4种情况。结合端元提取算法分类,针对算法缺陷及改进思路,介绍了常见端元提取算法PPI、N-FINDR、UOSP、VCA、ICA、NMF和AMEE研究进展。最后,结合解混理论进展和工程应用实际,从技术综合和性能优化的角度指出了端元提取算法的研究展望。An explanation of endmember based on physics,algebra and geometry is described. And a classifi- cation,with three categories,of endmember extraction algorithms based on algorithm design theory is pro- vided, namely, endmember extraction algorithms designed based on geometry, endmember extraction algo- rithms designed based on statistics and signal detection theory,and endmember extraction algorithms de- signed based on combination of spectral and spatial information. Furthmore,the category based on geome- try can be subdivided into four cgnditions according to the different techniques, that is, distance, volume, projection and transformation, optimization. Owing to the classification of endmember extraction algo- rithms,the defects and improved techniques, research progress of some commonly endmember extraction algorithms including PPI, N-Findr, UOSP, VCA, ICA, NMF, and AMEE are described. At last, from the point of view on engineering application of hyperspectral remote sensing and the development of unmixing theory, two research prospects on endmember extraction algorithm are pointed out. One prospect is combi- nation of all different techniques used in endmember extraction,and the other is the performance optimiza- tion of existing algorithms.

关 键 词：高光谱遥感 端元提取 线性光谱混合模型 性能优化 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]