遥感数据时域滤波与重建的谐波分析扩展方法研究

发布时间：2018-03-23 21:27

  本文选题：遥感数据　切入点：影像修复　出处：《武汉大学》2016年博士论文

【摘要】：随着卫星和传感器技术的不断创新,越来越多的遥感卫星观测数据得以获取。其中,遥感时间序列数据被广泛应用于区域以及全球环境变化研究中。随着研究的深度开展,对遥感时间序列数据时空连续性和完整性的要求越来越高。但是由于在数据获取的过程中,不可避免的会受到观测条件和传感器故障等因素的影响造成大量信息缺失,使得遥感时间序列数据呈现时间不连续、空间不完整状态,严重的阻碍了数据的进一步应用。如何在现有观测条件下,利用已有的遥感时间序列数据重建出高质量的时空连续的、完整的遥感时间序列数据促进了时域滤波和时域重建技术的发展。本文以提高遥感时间序列数据的时空连续性和完整性为主线,针对遥感时间序列数据产品中存在的问题和当下现有的时域滤波和时域重建方法的缺陷与不足,提出新的时域滤波和时域重建算法,以重建出高质量的遥感时间序列数据。本文主要工作总结为以下几方面：(1)以提高遥感时间序列数据的时间连续性和空间完整性为主线,分析了当下遥感时间序列数据产品存在的问题和遥感影像修复方法的研究现状,总结了它们在实际应用中存在的缺陷与不足,提出本文的研究目标。并且具体介绍了当前的时域滤波和时域重建算法以及论文中用到的评价方法与指标。(2)提出移动加权谐波分析的NDVI时域滤波方法。为了改善谐波分析方法(HANTS)重建结果出现过度拟合或过度平滑的现象,本文在原方法的基础上引入移动支持域,对时间序列数据进行移动加权局部处理。在每一个移动支持域中,通过三次样条法为每个数据点分配权值。在每个移动支持域中对数据进行拟合,并且通过权值分配控制参考数据对重建数据的影响程度：同时由于在移动支持域内待拟合数据少,使得谐波个数更加容易确定。此外,本文针对NDVI时间序列数据的特点设计四步处理流程对其进行处理,使得重建数据逼近原始NDVI的上包络线,从而更精确的获得植被的真实变化趋势。实验证明,该方法不仅可以很好的识别噪声点并且使得重建结果逼近NDVI时间序列的上包络线；还能够正确的估计植被休眠期的NDVI值,很好的处理NDVI时间序列中出现连续波动的现象,在绝大多数情况下鲁棒性强。(3)提出了谐波分析与泊松方程协同的地表反射率时域重建方法。针对已有方法无法有效的实现对每天数据的重建这一缺陷,本文提出的时域重建方法不仅可以重建每天的反射率时间序列数据,而且在保留未缺失区域原始值的前提下只对缺失区域进行填补,实现了真正意义的时域重建。主要思想是首先通过多年数据间的联合加权平均对待重建年份的缺失数据进行初步填补,为时间域重建提供足够的初始值；基于这些初始值通过时域滤波算法对第一步未填补的区域进行填补,同时对已经填补的区域进行调整；最后通过泊松图像编辑对修复区域的值进行调整,实现重建数据的空间无缝。实验结果表明该方法可以重建出每天的时空连续的地表反射率产品,不仅能够保持数据时域的连续性,也能保证数据空间的完整性,同时,重建的地表反射率数据也保持了光谱的完整性。(4)提出了顾及物理约束的地表温度时域重建方法。考虑到云层会对地表温度产生影响,本论文提出在地表温度数据的重建过程中充分考虑到云层对地表温度的影响,耦合能表现反映影响程度的物理量,通过物理约束手段重建出更符合真实情况的地表温度产品。该方法首先依据遥感时间序列产品时间上的依存性重建出晴空条件下高质量产品,然后建立物理约束辅助数据与待重建数据之间的关系,通过对关系模型参数进行高精度训练,将云覆盖区域的信息进行重建。实验结果表明该方法不仅可以提高晴空条件下质量低的像元的质量,而且可以对云覆盖区域的像元进行高精度的重建,最终重建出高质量的每天的地表温度数据。
[Abstract]:With the continuous innovation of satellite and sensor technology, remote sensing satellite data to get more and more. Among them, the remote sensing data of time series is widely used in regional and global environmental change research. With the depth of research carried out on the data of time series remote sensing, spatial continuity and integrity of the increasingly high demand. But in due process the data acquisition, will be influenced by the observation condition and sensor fault caused by a large number of factors such as lack of information, showing the time discontinuous remote sensing data of time series, space is not complete, seriously hinder the further application of data. How the existing observation condition, using the remote sensing data of time series reconstruction in time and space the high quality of the continuous, remote sensing time series data integrity and promote the development of time domain filtering and time domain reconstruction technology. In this paper. High spatial remote sensing data of time series continuity and integrity as the main line, aiming at the existing defects of remote sensing data products in time series and the existing time-domain filtering and time domain reconstruction method and the insufficiency, proposed time-domain filtering and time domain reconstruction algorithm of the remote sensing data of time series to reconstruct high quality. This paper the work summarized as follows: (1) to improve the remote sensing data of time series time continuity and spatial integrity as the main line, analyzes the research status of the existing remote sensing data of time series products and remote sensing image restoration method, summarizes their shortcomings in the practical application and the insufficiency, puts forward the research target in this paper. And introduces the specific evaluation methods and indicators used in the time domain and time domain filtering and reconstruction algorithm in this paper. (2) proposed moving weighted harmonic analysis ND VI time domain filtering method. In order to improve the harmonic analysis method (HANTS) reconstruction results over fitting or over smoothing phenomenon, this paper introduces the mobile support domain on the basis of the original method, the time series data of mobile weighted local processing. In each mobile support domain, through the three spline method for each data point distribution weights. In each mobile support for data fitting in the domain, and the distribution of weight control reference data impact on the reconstruction data. At the same time because in the mobile support domain to fit the data, the number of harmonic is more easily determined. In addition, the four step process to process it according to the characteristics of NDVI time sequence data design, making the reconstruction of data envelope approximation of the original NDVI, so as to obtain more accurate vegetation real trend. The experimental results show that this method not only can be very good On the envelope of noise and the reconstruction results approach NDVI time series; also can estimate the correct vegetation dormancy period NDVI, continuous wave phenomena appear NDVI time series well, in most cases robust. (3) proposed the surface reflectance time domain reconstruction method for harmonic coordination wave analysis and Poisson equation. Because the existing methods can not effectively realize the reconstruction of the defect data every day, the time domain reconstruction method proposed in this paper can not only reflectance time series data reconstruction every day, and keeping the original value under the missing regions not only to fill the missing regions, realize the true meaning of the time domain reconstruction the main idea is the lack of data. Firstly, combined with weighted data between the average years of reconstruction years were preliminarily treated filled, for the time domain reconstruction early enough Initial value; these initial values of the first step of filling the area filled by temporal filtering algorithm based on simultaneous adjustment on the fill area; finally, Poisson image editing to adjust the value of the repair area, realize the reconstruction of data space seamlessly. Experimental results show that this method can reconstruct the continuous time every day the surface reflectance products, not only can maintain the continuity of data in the time domain, but also to ensure the integrity of the data space, at the same time, the surface reflectance data reconstruction also maintain the integrity of the spectrum. (4) the surface temperature time domain reconstruction method and Gu physical constraints. Considering the clouds will affect the surface temperature. The paper proposes the reconstruction process of surface temperature data in full consideration of the effect of clouds on the surface temperature, the coupling can reflect the influence degree of physical quantity, through physical ca. Beam means to rebuild more in line with the surface temperature product in the real situation. The method based on remote sensing time series on the time dependence of the reconstruction of high quality products under the clear sky condition, and then the relationship between the establishment of physical constraints and auxiliary data to be reconstructed data, through high precision training parameters of the relationship model, will rebuild information the cloud coverage area. The experimental results show that this method can not only improve the quality of low quality of the pixels under the clear sky condition, but also can be used for high precision reconstruction of pixel cloud coverage, finally reconstruct the surface temperature data of high quality every day.

【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：P237
，

本文编号：1655225