基于高光谱遥感的城市水网水体提取研究和实现

发布时间：2018-04-25 19:05

本文选题：高光谱 + 水体信息提取　；参考：《山东大学》2014年硕士论文

【摘要】：高光谱数据具有光谱波段多、波段连续及图谱合一等特点,能提供丰富细致的地表信息,可以捕捉到内陆水体复杂多变的特征。论文在对高光谱遥感数据进行特征分析和预处理的基础上,研究和实现了河网城市的水体提取工作,并对相应的水体提取方法进行性能分析和评价。论文的主要工作如下： (1)分析研究城市内主要地物的理论光谱模型,主要包括：水体、植被、土壤、城区和阴影,总结各个地物的光谱反射特性,为水体信息专题提取做准备；详细分析了高光谱数据源的数据特征、描述模型和预处理工作,并在Linux环境下实现了遥感快视图的生成,使得遥感影像分析工作更加便捷。 (2)在南方城市河网密布,阴影等影响因子较多的情况下,利用机载AISA获取的高光谱遥感影像,通过分析实验区域不同水体环境,划分为5个包含水体信息的数据场景,通过分析不同场景中的水体提取影响因子,确定高光谱数据源中水体的特征波段,发现除近红外波段外,水体的光谱曲线在690nm-770nm波段表现出了与其他地物不同的变化趋势。 (3)研究现有的水体提取经典方法,包括：谱间关系法、归一化差异水体指数法(NDWI, Normal Differential Water Index)和支持向量机(SVM, Support Vector Machine);同时根据高光谱遥感数据的特点,选择适用于各个经典算法的高光谱数据的特征波段,实现多场景多方法的水体提取。根据场景分析得到的水体高光谱数据的特征波段,利用近红外阈值判断同特征波段构造的水体指数影像结合,构造基于高光谱水体特征波段的决策树方法；经实验验证,此算法可以提高水体提取的速度和准确度。 (4)通过对谱间关系法、NDWI、SVM和基于特征波段的决策树算法的水提结果精度进行分析,发现高光谱数据源的特征波段能很好的抑制非水体的信息,尤其是城市建筑的阴影,为以后城市制图、地表覆盖分类和水体信息的研究提取奠定了基础。
[Abstract]:Hyperspectral data are characterized by multiple spectral bands, continuous bands and the integration of spectra, which can provide rich and detailed surface information and capture the complex and changeable characteristics of inland water bodies. On the basis of feature analysis and pretreatment of hyperspectral remote sensing data, this paper studies and realizes the extraction of water body in river network city, and analyzes and evaluates the performance of the corresponding water extraction method. The main work of the thesis is as follows: 1) analyze and study the theoretical spectral models of the main features in the city, including: water body, vegetation, soil, urban area and shadow, summarize the spectral reflection characteristics of each feature, and prepare for the special topic extraction of water information; The data features of hyperspectral data source, description model and preprocessing are analyzed in detail, and the fast view of remote sensing is generated in Linux environment, which makes the analysis of remote sensing image more convenient. 2) in the case of dense river network in southern cities and more influence factors such as shadow, the hyperspectral remote sensing images obtained by airborne AISA are divided into five data scenes containing water information by analyzing different water environment in the experimental area. By analyzing the influence factors of water extraction in different scenes, the characteristic bands of water bodies in hyperspectral data sources were determined. It was found that the spectral curves of water bodies showed a different trend from other ground objects in the 690nm-770nm band except for the near infrared band. (3) to study the existing classical methods of water body extraction, including: spectral relationship method, normalized differential water body index method (NDWI), Normal Differential Water Index) and support vector machine (SVM), Support Vector machine, and according to the characteristics of hyperspectral remote sensing data, The feature bands of hyperspectral data suitable for each classical algorithm are selected to achieve multi-scene and multi-method water extraction. According to the characteristic band of the hyperspectral data obtained from scene analysis, the decision tree method based on the characteristic band of hyperspectral water body is constructed by combining the near infrared threshold judgment with the water body index image constructed in the characteristic band. This algorithm can improve the speed and accuracy of water extraction. 4) by analyzing the accuracy of water extraction result of NDWI SVM and decision tree algorithm based on feature band, it is found that the feature band of hyperspectral data source can restrain the information of non-water body, especially the shadow of urban buildings. It lays a foundation for urban mapping, classification of surface cover and extraction of water information.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP751

【相似文献】