基于编组拟合的合成孔径雷达图像线特征提取  

Extraction of linear features in SAR images using grouping and fitting

作　　者：陈思[1,2] 杨健[1] 宋小全[2] 

机构地区：[1]清华大学电子工程系,北京100084 [2]北京跟踪与通信技术研究所,北京100094

出　　处：《清华大学学报（自然科学版）》2011年第2期166-171,共6页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(40871157)

摘　　要：为了从合成孔径雷达(SAR)遥感图像中高效率地提取线特征、满足目标识别与场景分析等应用的需要,采用曲线结构基元取代通用的直线结构基元,通过像素编组以及曲线拟合提取并连接,使得复杂的全局优化简化为基于连接关系和总长度的聚类和筛选,借助低分辨率图像对高分辨率图像的掩码操作先主后次地提取不同宽度的线特征。应用所提出的方法在实际SAR图像上进行了实验,获得了与观察相一致、具有单像素宽度的线特征二值图像。实验结果表明:所提出的方法可以快速准确地提取出场景中真实的线特征,采用曲线基元形式和编组拟合方法有利于降低问题复杂度、提高处理效率。The efficiency of extracting linear features from synthetic aperture radar（SAR） remote sensing images for target recognition and scene analysis is improved by replacing the commonly used straight segments by curved segments.These are extracted and connected through pixel grouping and curve fitting,so that the complicated global optimization is simplified to clustering and screening based on the connection relationships and total lengths.The linear features of different widths are extracted using the first-primary-last-secondary method by mask processing from low resolution images to high resolution images.Test on real SAR images using this method identified single-pixel wide linear features in binary images consistent with observations.The actual linear features in the scene were extracted rapidly and accurately with this method.Use of the curved segments and the grouping-fitting method reduces the problem＇s complexity and improves the processing efficiency.

关 键 词：测绘雷达 线特征 像素编组 曲线拟合 多尺度 

分 类 号：TN959.3[电子电信—信号与信息处理]