基于单形体体积增长的高光谱图像端元提取及快速实现

发布时间：2018-02-15 15:39

本文关键词： 遥感端元提取单形体体积增长算法分块矩阵 Cholesky分解　出处：《浙江大学》2015年硕士论文　论文类型：学位论文

【摘要】：高光谱遥感数据以其波段多、光谱分辨率高、数据量大等特点而成为当前遥感领域的前沿技术,在各个领域发挥着越来越大的作用。但是由于地面物质类型的复杂性以及成像系统空间分辨率的限制,高光谱图像中普遍存在混合像元,因此光谱解混是遥感领域的重要研究方向。而端元提取作为光谱解混的关键步骤,如何有效而快速地进行端元提取是高光谱遥感图像处理的研究重点之一。本论文主要针对端元提取算法中比较常用的基于线性光谱混合模型的新的单形体体积增长算法NSGA中存在的主要问题进行了一系列的改进,不仅将其扩展至适用于非线性光谱混合模型,而且提出了两种思路来解决其高计算复杂度的问题。论文的主要工作如下： (1)针对NSGA只适用于线性光谱混合模型而无法应用于非线性光谱混合模型的问题,本文利用核函数的方法实现该算法的非线性扩展,提出适用于非线性光谱混合模型的算法KNSGA. (2)针对基于线性模型的NSGA和非线性模型的KNSGA两算法中由重复体积计算而造成的高计算复杂度的问题,利用分块矩阵的性质提出了两种快速实现算法FNSGA和FKNSGA.两种快速算法主要通过利用分块矩阵的性质,来简化单形体体积公式行列式求解过程,从而减小时间及运算复杂度,达到简化算法,缩短算法运行时间的目的。 (3)针对(2)中提到的NSGA和KNSGA中存在的高计算复杂度问题,利用改进的Cholesky分解的方法提出了两种相应的快速实现算法FNSGACF和FKNSGACF。两种快速算法主要利用改进Cholesky分解方法,将求解最大单形体体积的计算转化为寻找矩阵对角元素最大的过程,从而避免直接的体积计算,降低了计算复杂度,达到快速实现的目的。在上述改进思路的基础上,本文采用仿真数据实验和真实高光谱图像实验两部分实验来对本文提出的改进算法进行实验验证,实验结果表明扩展算法KNSGA能够准确有效地提取端元,并且四种快速算法也能在准确提取端元的前提下缩短运行时间,达到快速实现的目的。
[Abstract]:Hyperspectral remote sensing data has become the frontier technology in the field of remote sensing because of its many bands, high spectral resolution and large amount of data. However, due to the complexity of the type of material on the ground and the limitation of spatial resolution of imaging system, mixed pixels are widely used in hyperspectral images. Therefore, spectral demultiplexing is an important research direction in remote sensing field. End-element extraction is the key step of spectral unmixing. It is one of the key points in hyperspectral remote sensing image processing how to extract endcomponents efficiently and quickly. This paper mainly focuses on the new volume increase of single body based on linear spectral mixed model which is commonly used in End-component extraction algorithm. The main problems in the long algorithm NSGA are improved. It is not only extended to the nonlinear spectral mixing model, but also two ideas are proposed to solve the problem of high computational complexity. The main work of the thesis is as follows:. 1) aiming at the problem that NSGA can only be used in linear spectral mixing model but not in nonlinear spectral mixing model, the kernel function method is used to realize the nonlinear expansion of the algorithm, and a new algorithm, KNSGA, which is suitable for nonlinear spectral mixing model, is proposed in this paper. In order to solve the problem of high computational complexity caused by repeated volume calculation in NSGA algorithm based on linear model and KNSGA algorithm based on nonlinear model, Two fast algorithms, FNSGA and FKNSGA, are proposed by using the properties of block matrix. By using the properties of block matrix, the process of solving determinant of volume formula of single body is simplified, and the time and computational complexity are reduced. The purpose of simplifying the algorithm and shortening the running time of the algorithm is achieved. In order to solve the problem of high computational complexity in NSGA and KNSGA, two corresponding fast implementation algorithms, FNSGACF and FKNSGA CFS, are proposed by using the improved Cholesky decomposition method. The two fast algorithms mainly use the improved Cholesky decomposition method. The calculation of the maximum volume of a single body is transformed into the process of finding the largest diagonal element of the matrix, thus avoiding the direct volume calculation, reducing the computational complexity and achieving the goal of fast realization. On the basis of the above improved ideas, this paper uses two experiments, simulation data experiment and real hyperspectral image experiment, to verify the improved algorithm proposed in this paper. The experimental results show that the extended algorithm KNSGA can extract endelements accurately and effectively, and the four fast algorithms can shorten the running time and achieve the purpose of fast implementation.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP751

【参考文献】