基于对象的合成孔径雷达影像极化分解方法

发布时间：2018-04-27 06:16

本文选题：合成孔径雷达 + 对象　；参考：《计算机应用》2014年05期

【摘要】：面向对象方法已成为全极化合成孔径雷达(SAR)影像处理的常用方法,但是极化分解仍以组成对象的像素为计算单元,针对以像素为单位的极化分解效率低的问题,提出一种面向对象的极化分解方法。通过散射相似性系数加权迭代,获得对象的极化表征矩阵并对其收敛性进行了分析,以对象极化表征矩阵的极化分解代替对象区域内所有像素的分解,提高极化特征获取效率。在此基础上,综合影像对象空间特征,并通过特征选择与支持向量机(SVM)分类进行分析和评价。通过AIRSAR Flevoland影像数据实验表明,面向对象的分解方法能够减少对象极化特征提取的时间,同时提高地物目标的分类精度。相对于监督Wishart方法,提出方法的总体精度和Kappa值分别提高了17%和20%。
[Abstract]:Object-Oriented method has become a common method for image processing of fully polarized synthetic Aperture Radar (SAR), but polarization decomposition still takes the pixel of the object as the computing unit, aiming at the problem of low efficiency of polarimetric decomposition based on pixel. An object-oriented polarization decomposition method is proposed. Through the weighted iteration of scattering similarity coefficient, the polarization representation matrix of the object is obtained and its convergence is analyzed. The polarization decomposition of the object polarization representation matrix replaces the decomposition of all pixels in the object region to improve the efficiency of obtaining polarization features. On this basis, the spatial features of image objects are synthesized and analyzed and evaluated by feature selection and support vector machine (SVM) classification. The experiment of AIRSAR Flevoland image data shows that the object-oriented decomposition method can reduce the time of polarimetric feature extraction and improve the classification accuracy of objects. Compared with the supervised Wishart method, the total accuracy and Kappa value of the proposed method are increased by 17% and 20%, respectively.
【作者单位】：上海交通大学自动化系;系统控制与信息处理教育部重点实验室(上海交通大学);
【基金】：国家973计划项目(2012CB719903) 国家自然科学基金委创新研究群体资助项目(61221003) 国家自然科学基金资助项目(41071256);国家自然科学基金青年科学基金资助项目(41101386) 上海交通大学理工交叉基金资助项目(X198144)
【分类号】：TN958

【参考文献】