基于光谱解混和目标优化的高光谱图像亚像元定位研究

发布时间：2018-11-11 17:01

【摘要】：高光谱图像包含了丰富的光谱信息,普遍的应用于多种领域,逐渐成为了对地观测最为首要的信息源之一。但是由于其成像原理,以及高光谱成像仪的制造技术等诸多外界原因的限制,高光谱图像的空间分辨率普遍比较较低,混合像元普遍存在于图像中。对于土地覆盖制图、海岸线提取、变化检测和景观指数估计等应用来说,混合像元内地物的空间细节信息极其重要,假如按照传统的硬分类方法,图像中的混合像元被归类为任意一种地物都是不正确的。亚像元定位恰好是弥补上述不足的有效方法。因此,亚像元定位技术具有极其重要的研究意义。本文以高光谱图像的光谱解混为基础,结合智能优化算法,研究了高光谱图像亚像元定位。论文的主要工作包括：(1)简单描述了本文研究的相关背景与实际意义,查阅国内外研究相关的文献,并对其进行分析与总结,为本文提出改进的亚像元定位方法提供重要的科学参考与理论支持。(2)对光谱解混相关理论进行系统介绍,包括光谱解混的定义及其数学模型,然后介绍了纯像元假设下的光谱解混典型方法,最后对无纯像元假设下的光谱解混典型方法进行简单介绍。(3)基于光谱解混和目标优化的亚像元定位算法的总体框架,确定以图像连通区域周长最小为目标函数,并介绍了三种不同的图像周长计算方法,进一步分析了适用于亚像元定位的优化算法。为了减少算法的时间复杂度,根据地物空间分布特点,提出了—种新的目标优化迭代策略,采用局部分析代替全局分析。(4)分别阐述了遗传算法和二进制粒子群算法的基本原理,以及两种算法在亚像元定位中的具体应用,包括群体的表示及更新过程,并结合3种不同的目标函数计算方法,对比了两种优化算法在亚像元定位中的应用效果。(5)通过分析连通区域存在特例时基于链码长度求周长最小无法保证结果最优的原因,提出修改孤立区域的周长并考虑连通区域个数构造代价函数,最后利用二进制粒子群优化实现亚像元定位。(6)总结本文所做工作,并对未来高光谱图像亚像元定位方面的发展方向进行展望。
[Abstract]:Hyperspectral images contain rich spectral information and are widely used in many fields and become one of the most important sources of information for Earth observation. However, due to the limitation of its imaging principle and the manufacturing technology of hyperspectral imager, the spatial resolution of hyperspectral images is generally low, and mixed pixels are widely used in images. For applications such as land cover mapping, shoreline extraction, change detection and landscape index estimation, spatial details of mixed pixel hinterland are extremely important, if the traditional hard classification method is used, It is incorrect to classify the mixed pixels in the image as any kind of feature. Sub-pixel location is an effective method to make up for the above deficiencies. Therefore, sub-pixel positioning technology is of great significance. Based on the spectral de-mixing of hyperspectral images and intelligent optimization algorithm, sub-pixel localization of hyperspectral images is studied in this paper. The main work of this paper includes: (1) briefly describing the background and practical significance of this study, consulting the relevant literature at home and abroad, and analyzing and summarizing it. This paper provides an important scientific reference and theoretical support for the improved sub-pixel localization method. (2) the related theories of spectral unmixing are introduced systematically, including the definition of spectral unmixing and its mathematical model. Then the typical method of spectral unmixing under pure pixel assumption is introduced. Finally, the typical method of spectral unmixing based on pure pixel assumption is briefly introduced. (3) the general framework of sub-pixel localization algorithm based on spectral deconvolution optimization is presented. The minimum circumference of the connected region of the image is determined as the objective function, and three different methods of calculating the circumference of the image are introduced, and the optimization algorithm suitable for sub-pixel location is further analyzed. In order to reduce the time complexity of the algorithm, and based on the spatial distribution of objects in the base area, a new iterative strategy of target optimization is proposed. Local analysis is used to replace global analysis. (4) the basic principles of genetic algorithm and binary particle swarm optimization algorithm are described respectively, and the specific applications of the two algorithms in sub-pixel localization, including the process of population representation and updating, are discussed. Combined with three different methods of calculating objective function, The application results of two optimization algorithms in sub-pixel location are compared. (5) the reason that the minimum circumference based on chain code length can not guarantee the optimal result is obtained by analyzing the existence of special cases in connected region. In this paper, we propose to modify the perimeter of isolated regions and consider the number of connected regions to construct cost functions. Finally, binary particle swarm optimization (BPSO) is used to realize sub-pixel localization. (6) the work done in this paper is summarized. The future development of hyperspectral image sub-pixel localization is prospected.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】