摘要利用遥感数据进行土地利用和地表覆盖分类研究是遥感技术最主要的应用 领域之一。遥感分类可以便于对当地情况的了解，便于其他研究。比如可以通过 遥感图像分类获得土地利用图，植被覆盖图以及其他一些图件，进而将这些图件 做为下一步的基础图件，进行环境，土地利用，植树造林等方面的研究。目前已 经有了大量可见光遥感的分类研究，与可见光相比 SAR 能够全天候工作，由于 微波具有一定的穿透力，因此 SAR 用来检测隐蔽目标，比如在树林中，草丛中 的物体。和真实孔径雷达比较 SAR 方位分辨率要明显高于前者，分辨率与距离 没有关系，相同条件下，真实孔径雷达的信噪比和距离成反比。所以，SAR 能 达到比真实孔径雷达更高的信噪比。70813

本文将利用具有高空间分辨率的合成孔径雷达 COSMO-Skymed 遥感影像， 开展地表分类研究，从而获得地表土地利用覆盖信息，比较不同分类方法的效果， 以及不同地面目标在雷达影像上的特点。依据本研究区土地覆盖和土地利用类 型，结合目视解译,确定研究区包含的主要地物类型，如农田、居民地、水体、 裸地。 结合 ISODATA，最大似然，支持向量机，决策树等分类方法，对遥感 图像进行分割，获取对地表实体更具代表性的图像对象 ,以提取出更多的分类辅 助信息，从而实现对地物的精确分类。最大似然法总体精度达到了 67.59%，决 策树总体精度达到了 62.43%，支持向量机总体精度达到了 65.26%。都在一定程 度上满足了研究区内大面积范围地表覆盖类型精度要求。

毕业论文关键词：遥感分类  COSMO-Skymed SAR 

Classification of Surface Synthetic Aperture Radar Data Based on COSMO-Skymed

Abstract The use of remote sensing data on land use and land cover classification is one of the most important applications of remote sensing technology. Remote Sensing Classification may facilitate understanding of local conditions, to facilitate other studies. For example, can be obtained by remote sensing image classification of land use, vegetation cover maps and other maps, and then these maps as the basis for the next maps, environmental, land use studies, afforestation and so on. There is now a large number of optical remote sensing classification, compared with visible light SAR can work around the clock, since the microwave has a certain penetration, so SAR is used to detect hidden targets, such as in the woods, grass objects. And compare the real aperture radar, SAR azimuth resolution to be significantly higher than the former, resolution and distance does not matter, under the same conditions, and inversely proportional to the signal to noise ratio from the real aperture radar. Therefore, SAR can achieve higher than the real aperture radar signal to noise ratio.

This article will use COSMO-Skymed remote sensing images with high spatial resolution synthetic aperture radar, surface to carry out classification, thereby obtaining surface land use coverage information, the effect of different classification methods, as well as various ground targets on the radar image features. Based on land cover and land use types of the study area, combining visual interpretation to determine the type of the study area contains the main feature, such as farmland, residential areas, water, bare ground. Combined ISODATA, maximum likelihood classification vector machines, decision trees and other support, remote sensing image segmentation, surface entity to obtain a more representative image object to extract more free auxiliary information, enabling feature exact classification. Maximum likelihood method then the overall accuracy reached 67.59%, the overall accuracy of the decision tree 62.43%, SVM overall accuracy reached 65.26%, They were to some extent to meet the study area, a large area land cover type accuracy.