Hyper-Spectral Unmixing Algorithms Based On An Anti-Noise Model

The hyperspectral imaging system is a powerful technology for spectroscopic analysis of any particular object, or part of an object, within a scene of interest. In hyper-spectral image analysis, as well as in many other applications, due to the resolution limitation of the sensors and the variability of the ground surface, the observation of one pixel may contain several disparate substances, causing it to be a "mixed pixel." Hence, Spectral Unmixing (SU) is necessary and becomes a hot research field. Traditional unmixing models aim to minimize the additive noises and realize the SU. In this paper, we focus on the noises which are from the process of the imaging. To do that, we first present some questions based on the theoretical analysis of the traditional unmixing models, then propose our novel anti-noise unmxing model and describe the corresponding anti-noise unmixing method. Our main contributions are as follows:1Firstly, study the principle of hyperspectral imaging. Secondly, analyze the noises which are generated in the imaging process. Thirdly, summarize the problem of preprocessing problem which is before unmixing, then a further research is given on the three popular unmixing models and the corresponding unmixing algorithms. Finally, some well known evaluation clitia are introduced in this paper.2Traditional SU methods are usually based on the LMM or nonlinear mixture model (NLMM), in which only the additive noise is considered. However in hyper-spectral applications, the other mixed noise play important roles. In this paper, we propose an anti-noise model for hyper-spectral unmixing. In the anti-noise model, all noises are addressed.3It is well known that Nonnegative Matrix Factorization (NMF) has been widely used in blind SU, and the existing blind SU methods are usually based on Linear Mixture Model (LMM). However, theoretical analysis reveals that traditional LMM is a fundamental factor that impedes the improvement of blind SU. As a result, A Novel Blind Spectral Unmixing Method(NBSUM) is proposed, in which the Conjugate Gradient (CG) is employed to calculate end-member spectral and abundance. The proposed algorithm is able to overcome the shortcoming of traditional LMM and provides more accurate results than other state-of-the-art approaches, both synthetic and real hyper-spectral data experimental results demonstrate the efficacy of the proposed method.4We propose an anti-noise method for hyper-spectral unmixing. In the anti-noise method, all the noises are addressed. To deal with the problems faced by LMM or NLMM and to tackle the anti-noise model, the following techniques are applied:(1) an endmember dictionary is constructed firstly to initialize the solution,(2) an approximated L0norm constraint is employed to prune the dictionary and fulfill the sparse coding, and (3) the Itakura-Saito (IS) divergence, instead of the Square of Euclidean Distance (SED) divergence, is utilized to construct a novel optimization function. The experimental results on both synthetic and real hyper-spectral data sets demonstrate the efficacy of the proposed model and the corresponding method.