高光谱图像的光谱解混模型与算法研究

发布时间：2018-04-03 13:19

本文选题：高光谱图像解混　切入点：全变分模型　出处：《电子科技大学》2017年硕士论文

【摘要】：高光谱成像是将成像技术与光谱技术相结合的技术,是遥感应用中一个快速发展的领域。高光谱图像在军事目标辨别、远程控制、生物医学、食品安全以及环境监测等领域都有重要应用。但由于高光谱成像光谱仪空间分辨率较低,使得每个高光谱像元可能由多种不同物质的光谱混合构成,因此混合像元广泛存在于高光谱图像中。混合像元导致科研实践中一些应用分类不准确,因此对混合像元进行分解是高光谱遥感应用亟待解决的核心问题。本文中首先介绍了两种光谱混合模型:线性和非线性光谱混合模型。线性模型假设观察到的像元信号是所有的纯光谱信号的线性组合。与之相反,非线性模型则考虑到多种物质反射光之间的物理相互影响。其次,本文对高光谱图像解混的几种经典模型进行介绍。在这些模型中详细介绍了本文的对比模型全变分模型(SUnSAL-TV),该模型利用高光谱图像空间关系构建了对端元丰度的正则项,这使高光谱图像解混问题在数值结果和视觉效果上都有较大提升。但全变分模型的缺点是解混后丰度图中原平滑区域中伴有阶梯效应现象,视觉效果欠佳。本文采用重叠组稀疏全变分作为端元丰度正则项,并采用交替方向乘子法对模型进行求解,将原问题转化为一系列较易求解的子问题,进而得到原问题的全局解。在应用交替方向乘子法进行求解过程中,关于梯度域重叠组稀疏的子问题采用采用优化最小化方法进行求解。通过合成数据和真实数据的实验证明,采用本文提出的新方法处理后图像视觉效果和数值效果相比SUnSAL-TV方法有明显提升,并且可以有效减弱SUnSAL-TV模型的阶梯效应,使处理后丰度图更加平滑,视觉效果更佳。
[Abstract]:Hyperspectral imaging, which combines imaging technology with spectral technology, is a rapidly developing field in remote sensing applications.Hyperspectral images have important applications in military target identification, remote control, biomedicine, food safety and environmental monitoring.However, because of the low spatial resolution of hyperspectral imaging spectrometer, each hyperspectral pixel may be composed of multiple spectral mixtures of different substances, so mixed pixels are widely used in hyperspectral images.Mixed pixels lead to inaccurate classification of some applications in scientific research practice, so decomposition of mixed pixels is the core problem to be solved urgently in hyperspectral remote sensing applications.In this paper, we first introduce two kinds of spectral mixing models: linear and nonlinear spectral mixing models.The linear model assumes that the observed pixel signal is a linear combination of all pure spectral signals.In contrast, the nonlinear model takes into account the physical interaction between the reflected light of a variety of substances.Secondly, several classical models of hyperspectral image unmixing are introduced in this paper.In these models, the contrasting model, total variational model, SUnSAL-TVN, is introduced in detail. By using the spatial relation of hyperspectral images, the canonical terms of opposite-end Yuan Feng degree are constructed.This improves the numerical results and visual effects of hyperspectral image demultiplexing.However, the disadvantage of the total variational model is that there is a step effect in the original smooth region in the unmixed abundance map, and the visual effect is not good.In this paper, the sparse total variation is used as the regular term of abundance of the end element, and the alternating direction multiplier method is used to solve the model. The original problem is transformed into a series of subproblems which are easy to solve, and the global solution of the original problem is obtained.In the process of solving the problem using alternating direction multiplier method, the optimal minimization method is used to solve the sparse subproblem of overlapped groups in gradient domain.The experimental results of synthetic data and real data show that the visual effect and numerical effect of the new method proposed in this paper are much better than that of SUnSAL-TV method, and the step effect of SUnSAL-TV model can be effectively reduced.After processing, the abundance map is smoother and the visual effect is better.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】