Band selection of hyperspectral data with low-rank doubly stochastic matrix decomposition

In this article, a clustering-based band selection method is proposed to tackle the dimension reduction problem of hyperspectral data. The method is essentially based on low-rank doubly stochastic matrix decomposition, which is more stable than current low-rank approximation clustering methods. Experimental results show that the selected band subsets perform well in hyperspectral data classification problems.