Affinity Matrix Learning Via Nonnegative Matrix Factorization for Hyperspectral Imagery Clustering

In this article, we integrate the spatial-spectral information of hyperspectral image (HSI) samples into nonnegative matrix factorization (NMF) for affinity learning to address issue HSI clustering. This technique consists three main components: 1) oversegmentation computing spectral-spatial matrix; 2) NMF with guidance obtained and 3) density-based spectral clustering on final matrix. First, is oversegmented superpixels via entropy rate superpixel algorithm. The defined based class-consistency assumption all in each similar between adjacent superpixels. Second, NMF, used guide iterative process NMF. then weighted by low-dimensional subspace form Third, applied obtain maps. Experimental results public benchmark HSIs demonstrate that proposed method superior considered state-of-the-art baseline methods both computational cost accuracy.