In hyperspectral images, some spectral bands suffer from low signal-to-noise ratio due to noisy acquisition and atmospheric effects, thus requiring robust techniques for the unmixing problem. This paper presents a robust supervised spectral unmixing approach for hyperspectral images. The robustness is achieved by writing the unmixing problem as the maximization of the correntropy criterion subject to the most commonly used constraints. Two unmixing problems are derived: the first problem considers the fully-constrained unmixing, with both the non-negativity and sum-to-one constraints, while the second one deals with the non-negativity and the sparsity-promoting of the abundances. The corresponding optimization problems are solved efficiently using an alternating direction method of multipliers (ADMM) approach. Experiments on synthetic and real hyperspectral images validate the performance of the proposed algorithms for different scenarios, demonstrating that the correntropy-based unmixing is robust to outlier bands.
Index TermsCorrentropy, maximum correntropy estimation, alternating direction method of multipliers, hyperspectral image, unmixing problem. F. Zhu is with the A. Halimi is with the School of Engineering and Physical Sciences, Heriot-Watt University, U.K. (a.halimi@hw.ac.uk) P. Honeine is with the LITIS lab, ) 2 I. INTRODUCTION
SPectral unmixing is an essential issue in many disciplines, including signal and image processing, with a wide range of applications, such as classification, segmentation, material identification and target detection. Typically, a hyperspectral image corresponds to a scene taken at many continuous and narrow bands across a certain wavelength range; namely, each pixel is a spectrum. Assuming that each spectrum is a mixture of several pure materials, the unmixing problem consists in two tasks: (i) identifying these pure materials (the so-called endmembers); (ii) estimating their proportions (the so-called abundances) at each pixel [1]. In practice, these two steps can be performed either sequentially or simultaneously [2]. Wellknown endmember extraction algorithms include the pure-pixel-based ones, e.g., the vertex component analysis (VCA) [3] and the N-FINDR [4], as well as the minimum-volume-based ones, e.g., the minimum simplex analysis [5] and the minimum volume constrained nonnegative matrix factorization [6]. While the endmember extraction is relatively easy from geometry, the abundance estimation remains an open problem. Indeed, the abundances can be estimated using least-squares methods, geometric approaches [2], or by tackling recently-raised issues such as nonlinearity [7], [8]. In this paper, we consider the abundance estimation problem. The linear mixture model (LMM) is the most investigated over the past decades [6], [9], [10]. Its underlying premise is that each pixel/spectrum is a linear combination of the endmembers. To be physically interpretable, two constraints are often enforced in the estimation problem: the abundance non-negativity constraint (ANC) and the a...