【作者】 朱俊杰；

【导师】 郭华东；

【作者基本信息】 中国科学院研究生院（遥感应用研究所） ， 地图学与地理信息系统， 2005， 博士

【摘要】 数据融合和信息提取是高分辨率遥感图像的研究热点。目前,基于小波多尺度分析的象素级遥感数据融合受到了广泛的关注,信息提取方法的研究也由过去的波段组合等简单方法向着基于知识的信息提取的复杂方法发展。 高分辨率遥感图像,尤其是高分辨率SAR图像,它们反映了地物目标更加丰富的信息,它们的出现满足了人们对精确的地物目标信息获取的需求。这些高分辨率的遥感图像必然会被广泛的应用,因此相关关键技术的研究变得越来越迫切。如何提高多光谱图像的分辨率,使其更加真实的反映地表;如何从高分辨率遥感图像中精确、快速地提取地物目标的结构、位置信息等等,都是具有重要意义的研究方向。 基于以上研究热点和研究状况,本文针对高分辨率的卫星光学影像和高分辨率的机载SAR图像进行了分析研究。从成像机理等方面对高分辨率SAR图像中的典型目标进行了深入的分析之后,利用小波多尺度分析理论、纹理分析技术、基于目标成像知识的理论等等,对高分辨率遥感图像开展了数据融合和信息提取等方面的研究工作,阐述了高分辨率SAR图像的一些应用方向,并提出了可行的技术方法。 本文的主要创新点有如下几个方面: (1) 提出了一种保持图像光谱特征、提高图像空间分辨率的高分辨率光学遥感图像融合方法。根据小波理论和局部相关系数对北京中关村地区的快鸟图像的全波段和多光谱波段进行了融合,这种方法在增加遥感图像信息、提高空间分辨率的同时,其光谱特征能够得到有效的保持,解决了使用小波方法在提高图像空间分辨率的同时,如何抑制融合图像光谱畸变的问题。得到的融合图像能更真实的描述地表,可以用来更精确的制图、提取、反演等等。 (2) 提出了高分辨率SAR图像中去除建筑物阴影虚警的水体提取方法研究。在分析淮河洪水监测的高分辨率SAR图像的目标特征之后,利用了建筑物的纹理特征和成像知识对图像中的建筑物阴影进行了检测并将它们从黑斑中去除,完成了对复杂度高、干扰强、虚警率大的水体提取,得到了满意的效果。使用此方更多还原

【Abstract】 Data fusion and information extraction are the research hotspots in remote sensing fields. At present, the research on data fusion based on wavelet transform is to be more attached importance to. However, the information extraction now introduces the knowledge-based method, which is more complicated than some simple methods such as the band combination method.High-resolution remote sensing image, especially high-resolution SAR image reflects the abundant information of objects on the earth’s surface. The image can satisfy the need to obtain precise information of the objects. Therefore, the high-resolution remote image will be widely used and the relative key techniques need to be studied imminently. The method of improving the resolution to make the image really reflect the earth’s surface and the method of precisely and fastly extracting the objects’ information on structure and location information are the significative research orientations.With the above research hotspots and research status, this paper analyzed and researched on the high-resolution optical and SAR images. After analyzing some typical objects on the basis of the imaging mechanism of SAR image, we used the wavelet theory, texture analysis technique and knowledge-based method to study data fusion and information extraction from high-resolution remote sensing images. We pointed out some research orientations and gave the feasible methods.The innovations in this dissertation are as following:(1) This paper put forward a kind of fusion method, which could keep the spectral feature and improve the spatial resolution of high-resolution optical image. Based on the wavelet theory and local correlation coefficient, we fused the panchromatic band with the multi-spectral bands of Quickbird image of the Zhongguancun area in Beijing. This method has enhanced the information, improved the resolution and kept the spectral feature, and it resolved the problem of how to keep the spectral feature when fusing images to improve the resolution. The fused image can more really reflect the earth’s surface and be used in precise mapping, extraction and retrieval.(2) Water extraction was studied with detecting and deleting false alarms brought by the building shadows in high-resolution SAR image. After analyzing the object’s feature in high-resolution SAR image, which was acquired for monitoring the Huai river flood, we used the texture feature and imaging knowledge of building to detect and delete the buildings shadows, and extracted the water. The result was satisfying. This method can be used to estimate the rainfall and number of ponds.(3) The fitting method to extract the road from high-resolution SAR image was brought forward. The mode of SAR imaging and the shelter of some other objects to road make the road edge in SAR image unclear and irregular. Therefore, this paper introduced the approaches: extract the road skeleton, fit the skeleton, and build the centerline of road to finish road extraction from the high-resolution SAR image of the Zhongguancun area in Beijing. This method resolved the problem of how to extract the urban road from high-resolution SAR image. The method can provide the vector更多还原