面向大分辨率差异SAR图像的海冰漂移跟踪研究

发布时间：2018-02-01 20:32

本文关键词： 海冰漂移非同源数据超分辨率重建 SIFT 克里金插值　出处：《大连海事大学》2017年硕士论文　论文类型：学位论文

【摘要】：每年海冰的冻结及漂移对海上航运、海上生产等均有不同程度的影响甚至造成灾害,因此对海冰漂移进行近实时、全天候跟踪具有重要意义。高精度海冰漂移跟踪需要短时间间隔的高分辨率图像序列。同源星载SAR由于较长的过境周期,进而限制了海冰漂移跟踪的精度,而利用非同源SAR构建短时间间隔的图像序列,进行高精度海冰漂移跟踪研究是一个合理的解决方案。特征对应性和跟踪算法效率是影响海冰漂移跟踪准确性的两个重要因素。由于传感器不同,其波段、极化方式、分辨率、入射角、噪声水平等存在差异,导致非同源SAR之间有明显的特征差异。本文面向大分辨率差异非同源SAR图像的海冰漂移跟踪研究,主要工作和研究成果概括如下:首先,重点研究了非同源SAR特征对应问题。同源数据因为各参数相同所以对应海冰的特征相似,而非同源数据之间的参数差异,导致不同数据相同海冰存在特征上的差异,特别是在大分辨率差异的情况下,高分辨率呈现的一些细节特征在低分辨率中不存在一一对应关系,为此本文重点研究并提出了基于超分辨率重建技术,通过寻找稳定特征的方法来进行特征对应分析研究;其次,由于非同源数据之间存在明显的灰度差异,而一些经典的适用于同源数据海冰漂移跟踪算法(如MCC算法等)针对非同源数据常常会失效,为此本文设计并实现了一种基于SIFT算子的特征匹配及跟踪算法。该算法基于计算结构特征而非灰度特征,能够有效提高匹配及跟踪精度,并且SIFT算法具有的旋转不变性能够有效的解决海冰旋转问题;最后,因为海冰漂移过程中会出现挤压碰撞等情况,导致海冰漂移不是线性变化,本文使用了经典的非线性稳定内插值算法(普通克里金插值),并在此基础上对插值初始条件的设定进行了改进,最终实现了海冰漂移速度矢量场标定。
[Abstract]:Every year, the freezing and drift of sea ice have different degrees of influence on maritime navigation and production, and even cause disasters, so the drift of sea ice is carried out in near real time. High-precision sea ice drift tracking requires high resolution image sequences with short time intervals. Homologous spaceborne SAR has a long transit period which limits the accuracy of sea ice drift tracking. The non-homologous SAR is used to construct the short time interval image sequence. The research of high precision sea ice drift tracking is a reasonable solution. Feature correspondence and tracking algorithm efficiency are two important factors that affect the accuracy of sea ice drift tracking. The differences in polarization, resolution, incident angle, noise level and so on lead to obvious differences in characteristics between non-homologous SAR images. This paper focuses on the study of sea-ice drift tracking in non-homologous SAR images with large resolution differences. The main work and research results are summarized as follows: firstly, we focus on the problem of non-homologous SAR feature correspondence. Because of the same parameters, the corresponding sea ice features are similar. But the difference of parameters between the different data leads to the differences in the characteristics of the same sea ice, especially in the case of large resolution difference. Some detail features presented in high resolution do not have a one-to-one correspondence in low resolution. Therefore, this paper focuses on the super-resolution reconstruction technology. The method of finding the stable feature is used to analyze the correspondence of the feature. Secondly, because of the obvious grayscale difference between the non-homologous data, some classical algorithms (such as MCC algorithm) which are suitable for the sea-ice drift tracking of the homologous data, will often fail for the non-homologous data. In this paper, a feature matching and tracking algorithm based on SIFT operator is designed and implemented. The algorithm is based on computational structure features rather than gray features, which can effectively improve the accuracy of matching and tracking. And the rotation invariance of SIFT algorithm can effectively solve the problem of sea ice rotation. Finally, because the sea ice drift will be squeezed and collided, the sea ice drift is not a linear change. In this paper, the classical nonlinear stable interpolation algorithm (ordinary Kriging interpolation) is used. Finally, the calibration of ice drift velocity vector field is realized by improving the setting of interpolation initial conditions.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P731.15;TN957.52

【相似文献】