一种基于对抗学习的仿真遥感图像生成方法  

A simulated remote sensing image generation method based on adversarial learning

作　　者：马征 褚钧正 武鹏飞 MA Zheng;CHU Jun-zheng;WU Peng-fei(Troop 91245 of PLA,Beijing 100000;School of Statistics and Data Science,Nankai University,Tianjin 300071;Science and Technology on Special System Simulation Laboratory,Beijing Simulation Center,Beijing 100854,China)

机构地区：[1]中国人民解放军91245部队,北京100000 [2]南开大学统计与数据科学学院,天津300071 [3]北京仿真中心航天系统仿真重点实验室,北京100854

出　　处：《计算机工程与科学》2023年第3期489-494,共6页Computer Engineering & Science

基　　金：国家自然科学基金(62001252);国防科技重点实验室基金(61420020401)。

摘　　要：遥感图像数据标注耗时、成本高且需要专家知识,使得有标签的遥感数据难于获得,因此亟需生成有标签遥感数据的有效方法。由计算机视觉领域用于风格迁移的循环一致生成对抗网络出发,提出了一种基于深度学习,利用循环一致生成对抗网络生成新数据集的仿真遥感图像转换方法。将源数据与生成数据视为源域与目标域,遥感图像转换可视为仿真遥感数据集的风格迁移。生成的数据集可进一步用于分类、语义分割和域适应等适用于遥感图像的常见任务。实验结果表明该方法可有效生成风格迁移的仿真遥感数据。Remote sensing image data annotation is time-consuming and costly and requires expert knowledge,making it difficult to obtain remote sensing data with labels.Therefore,it is necessary to generate an effective method of remote sensing data with labels.Starting from the cycle-consistent generative adversarial networks for style transfer in the field of computer vision,a simulated remote sensing image conversion method based on deep learning and cycle-consistent generative adversarial networks to generate new dataset is proposed.This method regards the source data and the generated data as the source domain and the target domain,which can be regarded as the style transfer of the simulated remote sensing dataset.The generated dataset can be further used for common tasks of remote sensing images,such as classification,semantic segmentation,and domain adaptation.Experimental results show that this method can effectively generate simulated remote sensing data with style transfer.

关 键 词：仿真遥感 风格迁移 循环一致生成对抗网络 域适应 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]