Unsupervised Hyperspectral Band Selection Method Based on Low-Rank Representation

Abstract

In order to reduce the spectral redundancy of hyperspectral remote sensing images and reduce the computational complexity of subsequent processing, an unsupervised hyperspectral image band selection algorithm based on low-rank representation (LRBS) was proposed in this paper. First, a low-rank representation of the hyperspectral image is proposed and a low-rank coefficient matrix is obtained. Then, each column of the low-rank coefficient is used as a vertex of the graph to perform spectral clustering. Lastly, we use the fixed initial k-means cluster centers for clustering to get the salient band of each cluster. The experimental simulation results show that the bands selected by LRBS algorithm can improve the classification accuracy and have better performance than other methods.