基于线性模型的端元数目估计和光谱提取算法研究

发布时间：2018-06-19 13:49

本文选题：LSMM + 端元数目估计　；参考：《华中科技大学》2014年硕士论文

【摘要】：高光谱遥感技术的快速发展已经引起了科学界越来越多的关注并在社会各领域得到了广泛应用。由于成像光谱仪空间分辨率的有限性和自然界地物的复杂性，混合像元在高光谱图像中普遍存在，成为制约遥感技术发展的瓶颈。端元选取是混合像元分解最关键的一步，它包括端元数目估计和光谱提取。线性光谱混合模型（LinearSpectral Mixture Model, LSMM）具有物理意义明确，构模简单的特点，也是国内外研究最深入、被运用最多的模型。因此，，本文的研究工作基于线性模型，具体如下：综述了线性模型下混合像元分解过程中端元数目估计、光谱提取、丰度估计等方法。对端元数目估计和光谱提取的常用算法进行深入研究，提出相应的改进算法。对基于特征值极大似然函数分布的特征值极大似然（Eigenvalue LikelihoodMaximization, ELM）算法进行改进，提出了基于噪声估计和噪声白化处理的改进的特征值极大似然（Modified Eigenvalue Likelihood Maximization, MELM）算法，并针对ELM算法的结果表达提出了改进的似然函数表示方法。对基于凸面单体的顶点成分分析（Vertex Component Analysis, VCA）算法，提出了一种低时间复杂度的改进的顶点成分分析（Modified VertexComponent Analysis, MVCA）算法。在高光谱数据处理过程中，MVCA略去VCA算法中不必要的几个步骤，用斯密特正交化改进正交向量产生过程，避免了高复杂度的求伪逆运算。在模拟数据集和真实数据集上的实验表明：MELM算法相比ELM算法有更高的精确度和鲁棒性，但同时也增加了一定的时间复杂度。将MVCA算法和VCA算法分别从端元提取精确度和浮点计算量两个角度进行比较，结果表明MVCA算法能够在保证同样精确度的同时产生更小的浮点计算量。
[Abstract]:The rapid development of hyperspectral remote sensing technology has attracted more and more attention from the scientific community and has been widely used in various fields of society. Due to the limited spatial resolution of the imaging spectrometer and the complexity of the natural features, mixed pixels are widely used in hyperspectral images, which become the bottleneck of the development of remote sensing technology. End-element selection is the most important step in the decomposition of mixed pixels, which includes the estimation of the number of endelements and spectral extraction. Linear Spectral mixture Model (LSMMM) has the characteristics of clear physical meaning and simple modeling. It is also the most deeply studied and used model at home and abroad. Therefore, the research work in this paper is based on the linear model, as follows: the methods of endmember number estimation, spectral extraction and abundance estimation in the process of mixed pixel decomposition under the linear model are reviewed. In this paper, the common algorithms for the number estimation and spectral extraction are studied, and the corresponding improved algorithms are proposed. The Eigenvalue LikelihoodMaximation (Elm) algorithm based on the distribution of eigenvalue maximum likelihood function is improved. An improved eigenvalue maximum likelihood maximization (MELM) algorithm based on noise estimation and noise whitening is proposed. An improved likelihood function representation method is proposed for the result representation of ELM algorithm. For vertex component analysis (VCA) algorithm based on convex unit, an improved modified Vertex component Analysis (MVCA) algorithm with low time complexity is proposed. In the process of hyperspectral data processing, MVCA omits some unnecessary steps of VCA algorithm, and improves the orthogonal vector generation process by using Smith orthogonalization to avoid the high complexity of pseudo-inversion operation. Experiments on simulated data sets and real datasets show that compared with ELM algorithm, the weight MELM algorithm has higher accuracy and robustness, but also increases the time complexity. The MVCA algorithm and the VCA algorithm are compared from the point of view of precision and floating-point computation, respectively. The results show that MVCA algorithm can generate less floating-point computation at the same time as the same accuracy.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【相似文献】