基于极化SAR非监督分类的油膜厚度估算方法研究

发布时间：2018-05-03 08:30

本文选题：全极化合成孔径雷达 + 极化特征分解　；参考：《大连海事大学》2015年硕士论文

【摘要】：利用全极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,简称PolSAR)数据进行海面溢油监测是海洋遥感的新领域之一,全极化数据相对单极化数据,包含丰富的极化特征信息和纹理信息,并且具有高效性、实时性、不受时间、气候限制等优势,因此全极化SAR海面溢油厚度估算方法的研究具有重要意义。与海冰及其他地物信息相比,由于海风、海浪、及其自身的化学反应,海面溢油的变化具有很大的动态性,这都增加了研究的难度。在海面溢油中,轮廓与厚度信息是溢油量的体现。本文采用多特征融合策略设计分类器,考虑到极化特征间的相关性,使用马氏距离对模糊C均值聚类算法改进,进行油膜厚度估算。本文的研究思路主要包含以下几方面：首先,分析了海面溢油的极化散射特性,研究并比较能够用于海面溢油的油膜厚度估算的单极化特征。其次,根据实验室的海面溢油数据,以及分类研究,提出了基于多特征融合策略的分类器,用于油膜厚度估算。根据各特征向量在油膜厚度估算中占的比重不同,分配不同的特征权值,进行多特征融合；分类器的设计是在模糊C均值聚类算法的基础上进行的改进,在算法的预处理阶段,加入马氏距离,自动计算初始聚类中心,进行油膜厚度估算。最后,在本文的实验中使用的数据是RADARSAT-2全极化数据,都采集的是发生在墨西哥湾地区的不同时间段与时间点的石油泄漏事故的三景海面溢油数据。对这三景数据做测试,得出部分结果。
[Abstract]:Sea surface oil spill monitoring based on Polarimetric Synthetic Aperture Radar, data is one of the new fields of ocean remote sensing. And it has the advantages of high efficiency, real time, not limited by time and climate, so it is of great significance to study the method of estimating the oil spill thickness of fully polarized SAR sea surface. Compared with the information of sea ice and other features, because of the sea wind, wave and its own chemical reaction, the oil spill on the sea surface has a great dynamic change, which makes it more difficult to study. In the oil spill, the information of profile and thickness is the embodiment of oil spill. In this paper, multi-feature fusion strategy is used to design classifier. Considering the correlation between polarization features, Markov distance is used to improve fuzzy C-means clustering algorithm to estimate oil film thickness. The main research ideas of this paper are as follows: firstly, the polarimetric scattering characteristics of oil spills on the sea surface are analyzed, and the one-polarization characteristics which can be used to estimate the oil film thickness on the sea surface are studied and compared. Secondly, a classifier based on multi-feature fusion strategy is proposed to estimate the oil film thickness according to the oil spill data of the laboratory and the classification research. According to the different proportion of each feature vector in the estimation of oil film thickness, different characteristic weights are assigned to carry out multi-feature fusion. The classifier is designed on the basis of fuzzy C-means clustering algorithm, and in the preprocessing stage of the algorithm, The initial cluster center is calculated automatically by adding Markov distance to estimate the oil film thickness. Finally, the data used in this experiment are RADARSAT-2 full polarization data, which are collected from three sea spills of oil spill accidents in different time periods and time points in the Gulf of Mexico region. The data of the three scenes are tested and some results are obtained.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：X87

【参考文献】