余旺盛1,,
王鹏1,,
陈天平1,,
邹焕新2,
1.空军工程大学信息与导航学院 西安 710077
2.国防科技大学电子科学学院 长沙 410073
基金项目:国家自然科学基金(41601436, 61403414, 61703423)，陕西省自然科学基础研究计划(2018JM4029)
详细信息
作者简介:秦先祥(1986–)，男，广西人，空军工程大学信息与导航学院讲师，主要研究方向为SAR图像智能处理与分析。E-mail: qinxianxiang@126.com
余旺盛(1985–)，男，湖南人，空军工程大学信息与导航学院讲师，主要研究方向为计算机视觉与图像处理。E-mail: xing_fu_yu@sina.com
王鹏：王 鹏(1985–)，男，山西人，空军工程大学信息与导航学院副教授，硕士生导师，主要研究方向为信息融合处理与分布式协同控制。E-mail: blueking1985@hotmail.com
陈天平(1979–)，男，四川人，空军工程大学信息与导航学院讲师，主要研究方向为智能信息处理技术。E-mail: chentianping1979@163.com
邹焕新(1973–)，男，广东人，国防科技大学电子科学学院教授，硕士生导师，主要研究方向为SAR图像解译、多源信息融合、计算机视觉、图像处理、模式识别等。E-mail: hxzou2008@163.com
通讯作者:秦先祥 qinxianxiang@126.com
责任主编：王爽 Corresponding Editor: WANG Shuang中图分类号:TN958
计量
文章访问数:1206
HTML全文浏览量:511
PDF下载量:216
被引次数:0
出版历程
收稿日期:2020-05-13
修回日期:2020-06-26
网络出版日期:2020-07-05
Weakly Supervised Classification of PolSAR Images Based on Sample Refinement with Complex-Valued Convolutional Neural Network
QIN Xianxiang1,,,YU Wangsheng1,,
WANG Peng1,,
CHEN Tianping1,,
ZOU Huanxin2,
1. Information and Navigation College, Air Force Engineering University, Xi’an 710077, China
2. College of Electronic Science, National University of Defense Technology, Changsha 410073, China
Funds:The National Natural Science Foundation of China (41601436, 61403414, 61703423), The Natural Science Basic Research Plan in Shaanxi Province (2018JM4029)
More Information
Corresponding author:QIN Xianxiang, qinxianxiang@126.com
摘要
摘要:针对物体框标注样本包含大量异质成分的问题，该文提出了一种基于复值卷积神经网络(CV-CNN)样本精选的极化SAR(PolSAR)图像弱监督分类方法。该方法首先采用CV-CNN对物体框标注样本进行迭代精选，并同时训练出可直接用于分类的CV-CNN。然后利用所训练的CV-CNN完成极化SAR图像的分类。基于3幅实测极化SAR图像的实验结果表明，该文方法能够有效剔除异质样本，与采用原始物体框标注样本的传统全监督分类方法相比可以获得明显更优的分类结果，并且该方法采用CV-CNN比采用经典的支持矢量机(SVM)或Wishart分类器性能更优。
关键词:极化SAR/
弱监督分类/
复值卷积神经网络/
样本精选
Abstract:In this study, a weakly supervised classification method is proposed to classify the Polarimetric Synthetic Aperture Radar (PolSAR) images based on sample refinement using a Complex-Valued Convolutional Neural Network (CV-CNN) to solve the problem that the bounding-box labeled samples contain many heterogeneous components. First, CV-CNN is used for iteratively refining the bounding-box labeled samples, and the CV-CNN that can be used for direct classification is trained simultaneously. Then, the given PolSAR image is classified using the trained CV-CNN. The experimental results obtained using three actual PolSAR images demonstrate that the heterogeneous components can be effectively eliminated using the proposed method, obtaining significantly better classification results when compared with those obtained using the traditional fully supervised classification method in which original bounding-box labeled samples are used. Furthermore, the proposed method with CV-CNN is superior to those in which the classical Support Vector Machine(SVM) and Wishart classifier are used.
Key words:PolSAR/
Weakly supervised classification/
Complex-Valued Convolutional Neural Network(CV-CNN)/
Sample refinement
PDF全文下载地址:
https://plugin.sowise.cn/viewpdf/198_d7ee7731-39ff-467d-b083-71828ba0d494_R20062
