Kernel Spatial-Spectral Based Multi-View Low-Rank Sparse Sbuspace Clustering for Hyperspectral Imagery

Clustering is an unsupervised task, which groups samples into classes without training datasets. It is a challenging task, especially for hyperspectral image (HSI), which contains more information in spectral bands than regular images. In this paper, kernel spatial-spectral based multi-view low-rank sparse subspace clustering (k-SSMLC) algorithm is proposed for HSI clustering. In the generated multi-view dataset, spectral views are created by spectral partition, spatial view is created by morphological feature extraction, and PCA provides clean view with less noise. After the multi-view dataset is created, kernel multi-view low rank sparse subspace clustering is applied. Experiments results demonstrate that the clustering accuracy of the proposed k-SSMLC is better than other HSI clustering algorithms, such as sparse subspace clustering (SSC), low-rank sparse subspace clustering (LRSSC), spectral-spatial sparse subspace clustering (S4C), and regular SSMLC.