遥感地表水体信息提取渐进增强模型研究

发布时间：2018-04-08 14:23

本文选题：水体信息提取　切入点：归一化差异植被指数　出处：《华南农业大学》2016年硕士论文

【摘要】：随着经济的快速发展,城市化、工业化对水体、水系污染的环境问题越来越突出,严重威胁着社会可持续发展。水体信息的准确获取对水资源调查、河流综合治理、水利规划、洪涝旱灾监测以及灾害评估等领域具有重要意义。从卫星遥感影像中快速准确地提取水体信息已成为众多水资源调查与监测的一种重要手段。但目前已有方法中,对一些背景地物较为复杂繁多的地区提取水体信息时,单一的提取方法无法有效的抑制非水体地物,从而导致水体提取精度不高。这主要原因是单一的水体提取方法往往只针对一种地物进行抑制,其他容易混淆的地物仍然存在于被提取的结果中。同时,在不同地区以及不同时期的影像提取水体结果时,往往需要人为地进行阈值的筛选,不利于计算机大范围,多时相地对影像进行水体提取。如何达到这一目的,正是本文研究的主要内容。本文选取了背景地物较为复杂的珠三角入海口西岸地区为研究区,同时选取三个不同时相的Landsat TM/ETM+影像作为数据源,在对典型的遥感水体信息提取方法进行深入分析的基础上,提出一种新的水体信息提取模型----地表水体信息提取渐进增强模型。最后,运用目视判读及人机交互矢量化解译的方法对地表水体信息提取渐进增强模型的提取结果与其他典型的遥感水体信息提取结果进行精度评价。从数据的预处理,到地表水体信息提取渐进增强模型的建立展开了一系列的研究工作,主要内容及成果如下:(1)常规的水体信息提取指数法(归一化差异植被指数、归一化差异水体指数、修正归一化差异水体指数),是利用水体及其他地物在不同波段的光谱差异的特征,从而达到水体提取的目的。但是,归一化差异植被指数对植被与水体区分不明显;归一化差异水体指数提取的水体容易与建筑物混淆;修正归一化差异水体指数为三者中最优,但仍有部分水体信息夹带着山体阴影,提取结果不够精确。(2)鉴于以上三种水体信息提取方法存在问题,本文提出遥感地表水体信息提取渐进增强模型。渐进增强模型首先对三个指数(归一化差异植被指数、归一化差异水体指数、修正归一化差异水体指数)进行处理;然后通过阈值设定,逐一隔离出水体信息;再对隔离结果进行二值化处理;最后进一步对三个隔离结果进行增强式的叠加,渐进增强了水体的信息,使水体提取信息达到更高的准确率、更快的提取速度,并解决了三种水体提取方法中存在的问题。(3)通过渐进增强模型与三种常规方法的比较试验,结果表明:以陆地三大类水体为试验对象,在湖泊、鱼塘水体试验中渐进增强模型比三种常规的方法总体精度提高了13%以上、总体kappa系数提高了0.26以上,在河流水体试验中增强模型比三种常规的方法总体精度和总体kappa系数都有所提高。遥感地表水体信息渐进增强模型可应用理论的分割阈值,为海量遥感地表水体信息提取过程的全自动化提供支持。本文使用的TM和ETM空间分辨率较低,混合像元较多,影响水体提取精度,在以后研究中,将考虑混合像元分解来进一步解决此问题。
[Abstract]:With the rapid development of economy, the city, the industrialization of water pollution, water environmental problems are becoming increasingly prominent, a serious threat to the sustainable development of the society. The investigation of water resources to obtain accurate information of water, water conservancy planning comprehensive management of rivers, plays an important role of flood drought monitoring and disaster assessment. From satellite remote sensing images fast accurately extract water body information has become an important means of investigation and monitoring of many water resources. But the existing methods, some of the more complex background area extracting water body information extraction method, a single can not effectively suppress the non ground water, resulting in water extraction accuracy is not high. The main reason is the water the extraction method is often only for a single object to suppress other confusing features still exist in the extracted results. At the same time, in the The water extract the same area as well as images of different periods, often require the screening threshold artificially, is not conducive to a wide range of computer, for water extraction of multitemporal images. How to achieve this goal, it is the main content of this paper. This paper chooses the background features more complex in the Pearl River Delta estuary West Bank as the study area, and selecting three different Landsat TM/ETM+ images as the data source, in-depth analysis based on remote sensing of water typical information extraction methods, proposes a new model to extract the information of water - surface water information extraction model of progressive enhancement. Finally, using visual interpretation and interactive vector solution the method of information extraction model of progressive enhancement of surface water extraction results to evaluate the accuracy and other typical water remote sensing information extraction results from the data. The pretreatment to the surface water information extraction model of progressive enhancement has launched a series of research work, the main contents and results are as follows: (1) water information extraction index method (normalized difference vegetation index, normalized difference water index, modified normalized difference water index), is characterized by water and other objects the spectral differences of different bands, so as to achieve the purpose of water extraction. However, normalized difference vegetation index of vegetation and water is not obvious; normalized difference water index to extract the water easily and building confusion; modified normalized difference water index is the best of the three, but there are still some water information extraction with the shadow of the mountain, the result is not precise enough. (2) in view of the above three kinds of water information extraction method has the problem, in this paper the water information extraction of remote sensing model gradually progressive enhancement. In the first three to enhance the model index (normalized difference vegetation index, normalized difference water index, modified normalized difference water index) for processing; then the threshold set, one by one to isolate the water information; isolation results of binarization processing; finally three isolation results were enhanced superposition, progressive enhancement the water information, the water information extraction to achieve higher accuracy, faster extraction, and solve the problems in existing methods of three kinds of water extraction. (3) through the progressive enhancement test, model and three kinds of conventional methods. The results showed that three kinds of land to water as the test object, in Lake, fish pond water test in progressive enhancement model than the three conventional method improves the overall accuracy by more than 13%, the overall kappa coefficient is improved by more than 0.26 in the water body test in enhanced model than the three General methods for the overall accuracy and kappa coefficient are improved. The overall surface water remote sensing information model of progressive enhancement threshold theory, automatic extraction process to provide support for the remote sensing information of surface water. TM and ETM spatial resolution using low mixed pixels more, affecting the accuracy of extraction of water, in the future in the study, considering the mixed pixel decomposition to solve this problem.

【学位授予单位】：华南农业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：P237

【参考文献】