基于光谱库的空谱联合高光谱稀疏解混方法研究

发布时间：2018-03-16 15:25

本文选题：高光谱解混　切入点：交替方向乘子法　出处：《南京信息工程大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,高光谱遥感技术发展显著,由于其具有较高的光谱分辨率,高光谱遥感已广泛应用于环境监测、目标侦察、伪装识别等重要领域。然而因为光谱成像仪的空间分辨率较低和地表的复杂多样性,多种物质(端元)的混合像元通常存在于高光谱遥感图像中。为了提高后续应用的精度,如何从混合像元中分解与提取各种地物光谱(端元)并求得其对应比例(丰度系数)至关重要。本文基于已知光谱库的线性光谱混合模型,充分利用高光谱数据的空间相关性以及丰度系数的稀疏性,进一步提高高光谱混合像元的解混精度。论文主要研究内容包括以下两个方面:(1)提出一种基于协同稀疏和TV正则化的高光谱解混模型。该方法在基于已知光谱库的线性解混模型中,引入TV正则项对高光谱邻域像元间的局部相关性进行约束;同时,为了进一步刻画丰度系数的全局行稀疏特性,在模型中加入协同稀疏;最后采用交替方向乘子法求解模型。(2)提出一种基于l1-l2 (l1范数减l2范数)稀疏性与TV正则化的高光谱解混模型。该方法利用l1-l2范数能够比l1范数产生更强的稀疏性,进一步降低光谱库中光谱特征高相关性的影响,以及利用TV正则项刻画高光谱像元的空间邻域相关性,促进像元间的平滑性。此方法通过结合l1-l2稀疏性与TV正则项建立稀疏解混模型,并利用交替方向乘子法求解模型,从而提高解混精度。模拟高光谱数据的实验结果证明本文提出的方法能够比现有同类方法获得更精确的解混结果,同时真实高光谱数据的实验结果验证了本文方法的有效性。
[Abstract]:In recent years, the development of hyperspectral remote sensing technology is significant, because of its high spectral resolution, hyperspectral remote sensing has been widely used in environmental monitoring, target detection, recognition and other important areas of camouflage. However because of the complexity of the diversity of the spatial resolution of the imaging spectrometer and the lower surface, a variety of substances (endmembers) of the mixed pixel usually exist in hyperspectral remote sensing images. In order to improve the accuracy of the subsequent application, how to decompose and extract various spectral from mixed pixels (endmembers) and its corresponding proportion calculated (abundance coefficient) is very important. In this paper, the linear spectral mixture model based on the known spectral library, make full use of the sparsity of the spatial correlation of hyperspectral data and the abundance coefficient, further improve the unmixing accuracy of hyperspectral mixed pixels. The main contents of this thesis include the following two aspects: (1) a method based on sparse and collaborative TV High spectral unmixing model. The method in the linear Library Based on known spectral unmixing model, the introduction of TV regularization of local high spectral correlation between pixels neighborhood constraints; at the same time, in order to further characterize the global row sparsity abundance coefficient, join in collaborative sparse model; finally, by using the alternating direction method to solve the model. (2) based on L1-L2 (L1 minus L2 norm norm) hyperspectral Unmixing Model sparsity and TV regularization. The sparse method using L1-L2 norm to produce a stronger than the L1 norm, to further reduce the influence of spectral spectrum feature high correlation, and describe the high by using TV regularization spectral pixel spatial correlation, and promote the smooth pixels. This method by combining L1-L2 and TV regularization based sparse sparse Unmixing Model and solving the model by using the alternating direction method of multipliers, from The experimental results of simulated hyperspectral data show that the proposed method can get more accurate results than the existing methods. Meanwhile, the experimental results of real hyperspectral data verify the effectiveness of the proposed method.

【学位授予单位】：南京信息工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】