Gaussian Kernel Dynamic Similarity Matrix Based Sparse Subspace Clustering for Hyperspectral Images

Abstract

Sparse subspace clustering algorithm (SSC) is one of the efficient methods for hyperspectral imagery (HSI) segmentation. However, it does not consider the spectral information and space information. In this paper, we propose a novel method Gaussian Kernel Dynamic Similarity Matrix Based Sparse Subspace Clustering (briefly GKD) which constructs adjacency matrix using sparse representation coefficient matrix and similarity matrix. Firstly, the sparse representation coefficient matrix is obtained by using the sparse subspace clustering model, and the similarity between pixel points after PCA dimension reduction is calculated according to the Gaussian Kernel function. Then, the similarity matrix of the graph is constructed by sparse coefficient matrix and Gaussian similarity. Finally, the cluster result can be received by applying the spectral clustering to the similarity matrix. The data experiments conducted in several datasets illustrate the improvements over the current state-of-the-art.