.png){kind=icon}
返回
Subspace multi-regularized non-negative matrix factorization for hyperspectral unmixing
.png){kind=icon}
PRE
AI
DOI：10.1007/s10489-022-04121-y.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Hyperspectral unmixing (HU) is an important task in hyperspectral image (HSI) processing, which estimates endmembers and their corresponding abundances. Generally, the unmixing process of HSI can be approximated by a linear mixing model. Since each type of material only appears in a few pixels in HSI, sparse non-negative matrix factorization (NMF) and its extensions are regarded as the main unmixing methods. However, due to the lack of consideration of the spatial distribution in the local domain, most HU methods lose the inducing effect of spatial correlation in the decomposition process, and ignore the correlation of the real distribution of different materials in different images. In this paper, a novel NMF unmixing model is proposed, called SMRNMF, which learns multiple subspace structures from the original hyperspectral images and combines them into a sparse NMF framework to improve the performance of the model. Firstly, subspace clustering is embedded into the sparse NMF model. According to the spatial correlation of the original data, two similarity matrices are learned, which make full use of the local correlations between the pixels of the original data. Secondly, based on the self-expression characteristics of the data in the subspace, the global similarity pixel graph matrix is embedded into the model to construct a self-expression regularizer to improve the unmixing performance. Finally, a smoothing matrix is cleverly constructed and embedded in the model to overcome the adverse effects of noisy information in the abundance images. Experiments on several simulated and real HSI data sets show that our method has superior performance compared with the existing methods.
Keyword:
Hyperspectral unmixing
Nonnegative matrix factorization
Multi regularized
Subspace structure
期刊
IF:
3.5
论文数:
7.6K
被引数:
1.7W
机构
引用论文
分享收藏A tabu search approach to automated map generalisation一种用于自动化地图简化的禁忌搜索方法.png)
Proceedings of the 10th ACM international symposium on Advances in geographic information systems
Proceedings of the 10th ACM international symposium on Advances in geographic information systemsIF0
PREAI
分享收藏Splitting or Lumping? A Conservation Dilemma Exemplified by the Critically Endangered Dama Gazelle (Nanger dama)PLoS ONE
OAAI
PLoS ONEIF0
OAAI分享收藏Structure constrained nonnegative matrix factorization for pattern clustering and classificationNEUROCOMPUTING
NEUROCOMPUTINGIF6.5
PREAI
分享收藏
The changing epidemiology of paediatric brain tumours: a review from the Hospital for Sick Children
PREAI
分享收藏A Gaussian Mixture Model Representation of Endmember Variability in Hyperspectral Unmixing高光谱分解中端元变异性的高斯混合模型表示OAAI
OAAI分享收藏
查看更多内容