Advances in Hyperspectral Image Processing Techniques

16Spectral‐Spatial Robust Nonnegative Matrix Factorization for Hyperspectral Unmixing

Risheng Huang¹, Xiaorun Li¹, and Liaoying Zhao²

¹College of Electrical Engineering, Zhejiang University, Hangzhou, Zhejiang, China

²School of Computer Science, Hangzhou Dianzi University, Hangzhou, Zhejiang, China

16.1 Introduction

Hyperspectral data are imageries collected by hypersepctral sensors, which consist of hundreds of contiguous narrow spectral band images. Due to the limitation of hypersepctral sensors' spatial resolution and mixing effects of ground materials, the observed spectrum for a pixel is commonly mixed by several materials' spectra. Therefore, hyperspectral unmixing (HU) is an important problem for the exploiting of remotely sensed hyperspectral data. The task of HU can be divided into two main steps, endmember extraction and abundance estimating. For endmember extraction, a number of algorithms have been proposed and widely used to find endmembers, including pixel purity index (PPI) [1], N‐FINDR [2], vertex component analysis (VCA) [3], and the simplex growing algorithm (SGA) [4]. The relationships among these methods have been investigated in Chang et al. [5], Li and Chang [6], and Chang et al. [7]. When the endmembers have been extracted, the abundances can be estimated by solving the constrained optimization problem under constraints including abundance nonnegativity constraint (ANC), abundance sum‐to‐one constraint (ASC), and so on [8].

Since the pure pixels may ...