›› 2014, Vol. 34 ›› Issue (4): 46-.doi: 10.3780/j.issn.1000.758X.2014.04.007

• 研究探讨 • 上一篇    下一篇

基于核部分非负矩阵分解的亚像元级地物光谱分析

 崔建涛, 厉小润, 赵辽英   

  1. (1浙江大学电气工程学院,杭州310027)(2杭州电子科技大学计算机应用技术研究所,杭州310018)
  • 出版日期:2014-08-25 发布日期:2014-08-25
  • 基金资助:

    国家自然科学基金(61171152),浙江省自然科学基金(LY13F020044),教育部支撑计划(625010216)资助项目

SpectralAnalysisforSubpixelMaterialsBasedonKernelPartialNonnegativeMatrixFactorization

 CUI  Jian-Tao, LI  Xiao-Run, ZHAO  Liao-Ying   

  1. (1CollegeofElectricalEngineering,ZhejiangUniversity,Hangzhou310027)(2InstituteofComputerApplicationTechnology,HangzhouDianziUniversity,Hangzhou310018)
  • Online:2014-08-25 Published:2014-08-25

摘要: 为了进一步提高亚像元级地物的光谱分析精度,提出了一种基于核部分非负矩阵分解KernelProtectionNonnegativeMatrixFactorization,KPNMF)的非线性解混算法。首先通过基于凸面几何理论的端元提取方法提取纯像元端元候选像素集合,然后根据候选像素的空间纯度指数判断纯像元端元。在纯像元端元信息已知的条件下,利用核方法对部分非负矩阵分解(ProtectionNonnegativeMatrixFactorization,PNMF)进行推广,构造相应的目标函数,推导迭代求解过程,分解求得亚像元端元光谱和所有端元的丰度。试验结果表明,提出的解混算法具有良好的非线性分解能力,解混结果优于线性解混算法。

关键词: 高光谱解混, 亚像元, 凸面几何, 空间纯度指数, 部分非负矩阵分解, 航天遥感

Abstract: Toimprovetheaccuracyofspectralanalysisforsubpixelmaterialsfurther,anonlinearunmixingalgorithmbasedonthekernelpartialnonnegativematrixfactorization(KPNMF)wasproposed.Firstly,anendmemberextractionalgorithmbasedonthetheoryofconvexgeometrywasusedtogenerateacandidatepixelsetofpureendmembers,andthenthepureendmemberwasdeterminedaccordingtothespatialpurityindicesofthecandidatepixels.Giveninformationofpureendmembers,thekernelmethodwasadoptedtoextendpartialnonnegativematrixfactorization(PNMF).Thecorrespondingobjectivefunctionwasconstructed,andtheiterativesolutionwasalsoderivedtoobtainthesubpixelendmembersandabundancesofalltheendmembers.Theexperimentalresultsdemonstratethattheproposedunmixingalgorithmhasgoodnonlinearunmixingability,andtheunmixingresultsarebetterthanthoseoflinearunmixingalgorithms.

Key words: Hyperspectralunmixing, Subpixel, Convexgeometry, Spatialpurityindex, Partialnonnegativematrixfactorization, Spaceremotesensing