首页期刊视频编委会征稿启事出版道德声明审稿流程读者订阅论文查重联系我们English
引用本文
  • 陈显,冯林,陈兵.融合云畸变和生成对抗网络的遥感影像云去除[J].电讯技术,2025,65(7):1110 - 1119.    [点击复制]
  • CHEN Xian,FENG Lin,CHEN Bing.Remote Sensing Image Cloud Removal by Integrating Cloud Distortion and Generative Adversarial Networks[J].,2025,65(7):1110 - 1119.   [点击复制]
【打印本页】 【下载PDF全文】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 254次   下载 124 本文二维码信息
码上扫一扫!
融合云畸变和生成对抗网络的遥感影像云去除
陈显,冯林,陈兵
0
(1.四川工商学院 计算机学院,成都 611745;2.四川师范大学 计算机科学学院,成都 610066;3.吉利学院 教务处,成都 641423)
摘要:
为解决光学遥感影像中云对图像的污染和遮挡问题,提出了一种融合云畸变和生成对抗网络的遥感影像云去除方法。首先,考虑云的透射、反射和吸收率变化,构建了云畸变物理模型。其次,将云畸变物理模型与生成对抗网络相结合,构建了遥感影像云去除模型,将输入图像分解为无云背景层和云畸变层,对输入突降进行重建,从而去除来自不同区域的未配对图像的云。最后,采用真实光学遥感图像对所提方法进行了实验验证,并与另外两种常规的云去除方法进行了对比分析。实验结果表明,所提方法在去除云层后,生成的无云图像的相关性R2值显著提高,分别从0.745、0.853、0.886上升至0.917、0.924、0.921,表明其在视觉和定量评估上均优于传统方法。
关键词:  遥感影像  云去除  云畸变  生成对抗网络(GAN)
DOI:10.20079/j.issn.1001-893x.240809005
基金项目:国家自然科学基金资助项目(52207112)
Remote Sensing Image Cloud Removal by Integrating Cloud Distortion and Generative Adversarial Networks
CHEN Xian,FENG Lin,CHEN Bing
(1.School of Computer Science,Sichuan Technology and Business University,Chengdu 611745,China;2.College of Computer Science,Sichuan Normal University,Chengdu 610066,China;3.Academic Affairs Office,Geely College,Chengdu 641423,China)
Abstract:
To solve the problem of cloud pollution and occlusion in optical remote sensing images,a cloud removal method for remote sensing images based on cloud distortion and generative adversarial network(GAN) is proposed.Firstly,the cloud distortion physical model is constructed by considering the transmission,reflection and absorption rate changes of cloud.Secondly,the cloud distortion physical model is combined with a GAN to build a remote sensing image cloud removal model.The input image is decomposed into a cloud-free background layer and a cloud distortion layer,and the input sudden drop is reconstructed,thus removing clouds from unmatched images from different regions.Finally,real optical remote sensing images are used to verify the proposed method,and the proposed method is compared with other two conventional cloud removal methods.The experimental results show that the correlation R-squared value of the cloud-free image generated by the proposed method is significantly improved after removing the cloud,from 0.745,0.853,0.886 to 0.917,0.924,0.921,respectively,indicating that the proposed method is superior to the traditional methods in visual and quantitative evaluation.
Key words:  remote sensing image  cloud removal  cloud distortion  generative adversarial network(GAN)
安全联盟站长平台