高分辨率SAR船只样本集构建与特征分析  

Construction and feature analysis of high resolution SAR ship sample set

作　　者：包萌 张杰 孟俊敏 张晰 郎海涛[3] BAO Meng;ZHANG Jie;MENG Junmin;ZHANG Xi;LANG Haitao(First Institute of Oceanography,Ministry of Natural Resources,Qingdao 266061,China;Ocean Telemetry Technology Innovation Center,Ministry of Natural Resources,Qingdao 266061,China;Beijing University of Chemical Technology,Beijing 100029,China)

机构地区：[1]自然资源部第一海洋研究所,青岛266061 [2]自然资源部海洋遥测技术创新中心,青岛266061 [3]北京化工大学,北京100029

出　　处：《电波科学学报》2019年第6期789-797,共9页Chinese Journal of Radio Science

基　　金：国家重点研发计划(2017YFC1405204);海洋公益性科研专项(201505002)

摘　　要：随着高分辨率合成孔径雷达(synthetic aperture radar,SAR)技术的不断发展,船只类型识别已成为遥感领域的重要研究课题.为满足在大样本支撑下的船只类型精确识别,文章利用RADARSAT-2和中国高分3号(GF-3)SAR数据构建了名为HR4S的高分辨率SAR船只样本集,详细阐述了构建HR4S的方法,并建立了一套完整的船只样本提取流程.该样本集涵盖1962个不同极化方式、分辨率以及类型的船只样本,在此基础上开展了船只几何参数分析,以及不同分类器与特征组合的船只类型识别性能分析等方面工作.结果表明:RADARSAT-2在HH、VH、VV极化中提取的几何参数均优于GF-3,并且航向在VV极化对船只几何提取影响最小;在类型识别性能上,随机森林(random forest,RF)分类器对GF-3船只分类精度最优达到了61.85%,而对于RADARSAT-2的船只分类精度最优达到了60.80%,GF-3船只分类精度优于RADARSAT-2.本文所构建的HR4S不仅进一步完善了高分辨率船只样本,并且在海上船只类型识别等方面具有的重要意义.With the development of high resolution synthetic aperture radar(SAR)technology,ship type recognition become smore and more important in remote sensing.In order to improve the identification accuracy,a high-resolution SAR ship sample set,named as HR4S,is constructed using RADARSAT-2 and Chinese GaoFen-3(GF-3)SAR data.The process of ship samples extraction and HR4S construction are introduced in detail.The HR4S covers 1962 samples with different polarization modes,resolutions and ship types.The ship geometry parameters and the ship classification performance of HR4S with different classifier and features are analyzed.The results indicate that the geometrical parameters extracted fromRADARSAT-2in HH,VH and VV polarization are all better than that of GF-3.Furthermore,the direction has little influence on the geometric parameter of ships in VV polarization.In terms of ship type recognition performance,the accuracy of random forest(RF)classifier achieved 61.85%on GF-3 data and 60.80%on RADARSAT-2 data.In general,the classification accuracy of GF-3 ships is better than RADARSAT-2.The HR4S constructed in this paper not only further improves the high-resolution ship samples,but also has important significance in the recognition of ship types at sea.

关 键 词：高分辨率SAR船只样本集 合成孔径雷达 船只几何参数 船只类型识别 

分 类 号：TN957.52[电子电信—信号与信息处理]