高分辨率遥感影像融合及面向对象信息提取研究

发布时间：2018-03-25 06:04

本文选题：高分二号　切入点：图像融合　出处：《北京交通大学》2017年硕士论文

【摘要】：高分二号卫星的成功发射标志着中国遥感卫星进入亚米级"高分时代"。到目前为止高分二号卫星发射仅仅两年多,针对高分二号影像的应用研究具有一定的意义。因此本文基于高分二号影像对其进行辐射定标、大气校正、正射校正、图像融合等预处理操作,其中主要研究了图像融合算法及融合后质量评价,并选取其中较优的融合影像进行后期的面向对象的覆被信息提取研究,为进一步深入研究奠定了一定的基础。本文的主要研究工作如下:(1)研究高分二号影像的预处理工作,系统的分析、总结了遥感影像数据融合的基本理论、特点和应用,深入研究分析了常用的9种融合算法,其中侧重研究了在实际应用中对遥感影像融合算法的选择问题。(2)针对生产过程中主要依赖主观人眼评价融合后影像,而主观评价与观察者的人眼视觉感知和经验水平相关性很大,不具备客观性,为此研究遥感影像客观评价指标和体系。但是在质量评价中即使相同的融合算法针对同一幅影像,应用者感兴趣的部分不同,影像的应用方向不同,对融合后质量的评价也不相同,需要针对具体影像做出具体评价,为此基于GDAL库利用c#语言编写客观指标参数计算软件,该软件将常用的指标参数集成在一起,为遥感影像质量评价提供可参考的、稳定的客观定量标准。最后得到本文实验数据较优的融合算法为Subtractive Resolution Merge。(3)以北京市通州区为实验区域,采用Subtractive Resolution Merge融合方法融合影像,利用面向对象分类方法完成对通州西北部主要城区大范围的覆被信息提取。主要研究了面向对象多尺度分割、分层提取的方法,研究了植被、水体、道路、建筑物在高分辨率影像上的特征,以此为基础建立规则集完成对研究区的分类,并与监督分类中标准最邻近分类法进行对比,结果表明,基于规则的面向对象分类结果精度相对于监督分类中标准最邻近分类结果精度较好。
[Abstract]:The successful launch of the Gaofen II satellite marks the entry of the Chinese remote sensing satellite into an "era of high scores" at the Yami level. So far, the launch of the Gaofen II satellite has only been more than two years old. Therefore, this paper carries on the radiometric calibration, atmospheric correction, orthophoto correction, image fusion and other preprocessing operations based on the high score 2 image. Among them, the image fusion algorithm and the quality evaluation after fusion are mainly studied, and the better fusion image is selected for the later research of object oriented covering information extraction. The main research work of this paper is as follows: 1) studying the preprocessing and systematic analysis of Gaofen II image, summarizing the basic theory, characteristics and applications of remote sensing image data fusion. In this paper, 9 common fusion algorithms are deeply studied and analyzed, in which the selection of remote sensing image fusion algorithm in practical application is mainly studied. In the process of production, it mainly depends on subjective human eyes to evaluate the fusion image. But the subjective evaluation has a great correlation with the human visual perception and experience level of the observer. Therefore, the objective evaluation index and system of remote sensing image are studied. But in the quality evaluation, even if the same fusion algorithm is applied to the same image, Different applications are interested in different parts, the application direction of images is different, and the evaluation of quality after fusion is also different, so it is necessary to make specific evaluation for specific images. Therefore, based on GDAL library, the objective index parameter calculation software is compiled by using c# language. The software integrates the commonly used index parameters together, and provides a reference and stable objective quantitative standard for the quality evaluation of remote sensing images. Finally, it is concluded that the best fusion algorithm for the experimental data in this paper is Subtractive Resolution Merge.Y3) the Tongzhou District of Beijing is taken as the experimental area. The Subtractive Resolution Merge fusion method is used to fuse the image, and the object oriented classification method is used to extract the covering information from the main urban areas in the northwest of Tongzhou. The method of object oriented multi-scale segmentation and hierarchical extraction is studied, and the vegetation is studied. Based on the characteristics of water bodies, roads and buildings in high-resolution images, a rule set is established to complete the classification of the study area and to compare it with the standard nearest neighbor classification in supervised classification. The results show that, The result precision of object oriented classification based on rules is better than that of standard nearest neighbor classification in supervised classification.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P237

【参考文献】