【作者】 马一薇；

【导师】 冯伍法；

【作者基本信息】 解放军信息工程大学 ， 摄影测量与遥感， 2010， 硕士

【摘要】 高光谱图像具有地物属性精细探测的能力和广阔的应用前景,高光谱图像处理与分析技术的研究正成为热点问题,但由于高光谱图像空间分辨率的提高受到限制,只有将其与高分辨率的遥感图像进行融合处理才能满足空间定位和属性获取的要求。本文在分析高光谱与高空间分辨率图像融合技术研究现状的基础上,深入研究了像素级融合方法,改进了非负矩阵分解融合方法,设计了基于遗传算法的融合方法,并对融合结果的质量评价方法进行了深入研究。主要完成的工作有:1.归纳了高光谱图像数据特点,分析了遥感图像融合的研究现状,强调了高光谱与高分辨率遥感图像融合的重要性。2.深入研究了像素级图像融合方法,介绍了高光谱图像预处理的内容和方法,阐述了图像融合中基于信息量和基于类间可分性的高光谱图像波段选择方法。在对多种像素级融合方法进行研究的基础上,分析了各种算法的实现方法并归纳总结了各自的优点和不足。3.分析了已有的遥感图像融合质量主、客观评价方法。为了实现综合评价的目的,将模糊综合评价应用到遥感图像融合的质量评价中,设计了基于层次分析法的融合图像质量模糊综合评价方法。试验结果表明:该方法具有兼顾主、客观因素,易于定量表达和区分优劣,评价结果精确等特点。4.在分析非负矩阵分解原理和实现方法的基础上,将IHS变换与非负矩阵分解方法相结合,改进了基于非负矩阵分解的图像融合方法。通过利用多光谱和高光谱数据进行试验,表明本文方法相对常用的融合方法具有克服光谱失真和保持图像信息量的优点,融合效果较好。5.对遗传算法的原理、流程和构成要素进行了研究,设计了基于遗传算法的图像融合方法。通过进行对比试验,分别对信息熵、相关系数和平均梯度等评价指标进行统计分析,表明了该方法具有方法灵活、信息互补优势明显的特点。更多还原

【Abstract】 Hyperspectral image is provided with the ability of detecting ground features accurately, as well as wide application foreground. The study on technique of processing and analyzing hyperspectral image is becoming a pop problem. However, the spatial resolution of hyperspectral image is restricted with being promoted, so only when fused with high spatial image, hyperspectral image could meet the needs for spatial position and attributes acquisition. Based on analyzing the study status of hyperspectrual image and high spatial image fusion, this dissertation lucubrates the method of fusion in the pixel level, improves the method of Non-negative Matrix Factorization (NMF) image fusion, devises the method of image fusion based on Genetic Algorithm (GA) and lucubrates the method of quality evaluation of fused images. The major works implemented are listed as follow:1.The characteristics of hyperspectral image data are summarized, the study status of remote sensing image fusion is analyzed and the importance of hyperspectral image fusion with the high spatial image is emphasized.2.Fusion methods in the pixel level are lucubrated, the matters and methods of preprocessing hyperspectral image are introduced, and the methods of hyperspectral image band selection based on information and classes distinguishability respectively are expatiated. On the basis that manifold methods of fusion in the pixel level are studied, implementation methods of these arithmetic are analyzed and virtue and defect of them are concluded.3.The existing subjective and objective quality evaluation methods of remote sensing image fusion are analyzed. In order to achieve the goal of comprehensive evaluation, fuzzy comprehensive evaluation is applied to the remote sensing image fusion quality evaluation and fuzzy comprehensive evaluation method of fusion images based on Analytic Hierarchy Process (AHP) is devised. Experiments show that it posses the characteristic of giving attention to subjective and objective factors, expressing in fix quantity and distinguishing virtues and defects easily, as well as evaluating accurately.4.Based on analysis of the principle and implementation of Non-negative Matrix Factorization (NMF), the image fusion method combined with IHS transform based on NMF is improved. Experiments on the multispectral and hyperspectral data indicate that the method in this dissertation possess virtues of getting over spectrum distortion and preserving image information, compared with the fusion methods in common use.5.The principle, flow and inscape of Genetic Algorithm(GA) are studied, the fusion method based on GA is devised. Contrast experiments show that the method is provided with characteristics of agility and obvious advantage of complementing information for each other.更多还原