摘要
在获取遥感图像的过程中,受天气等因素影响,得到的图像会含有云层,这直接影响了后期对于遥感图像的使用。针对这种问题,基于改进的生成对抗网络,提出了一种遥感图像去云方法。首先,网络的生成器主要结构为深度残差收缩网络,可以更好地去除噪声,并将生成器网络中的批归一化更换为组归一化以提高模型训练效率;其次,在网络损失函数中加入感知损失以进一步提高网络的去云效果。实验结果表明,相较于传统方法和深度学习方法,该方法在薄云处理时峰值信噪比最低提高了0.14 dB,最高提高了9.41 dB,结构相似性最低提高了0.01,最高提高了0.17。在厚云光学遥感图像去云处理方面,PSNR最高可提升6.97 dB,SSIM最高可提升到0.11,在主观视觉效果上也取得了较好的效果,验证了该方法的可行性和良好性能。
In the process of acquiring remote sensing images,affected by weather and other factors,the obtained images will contain clouds,which directly affects the later use of remote sensing images.To solve this problem,a remote sensing image cloud removal method is proposed,which is based on improved generative adversarial network.Firstly,the main structure of the generator is deep residual shrinkage network,which can better remove noise,and the batch normalization in the generator network is replaced by group normalization to improve the model training efficiency.Secondly,the perceived loss is added to the network loss function to further improve the cloud removal effect of the network.The experimental results show that compared with traditional methods and deep learning methods,the proposed method can improve the peak signal-to-noise ratio(PSNR)by 0.14 dB and 9.41 dB,and the structural similarity by 0.01 and 0.17 respectively in thin cloud processing.In the cloud removal processing of thick cloud optical remote sensing images,the maximum PSNR can be increased by 6.97 dB and SSIM can be increased to 0.11,and the subjective visual effect is also achieved,which verifies the feasibility and good performance of the proposed method.
作者
赵文翔
普运伟
ZHAO Wenxiang;PU Yunwei(Faculty of Land Resources Engineering,Kunming University of Science and Technology,Kunming 650093,China;Computer Center,Kunming University of Science and Technology,Kunming 650500,China)
出处
《遥感信息》
CSCD
北大核心
2024年第5期78-85,共8页
Remote Sensing Information
关键词
去云
生成对抗网络
组归一化
深度残差收缩网络
损失函数
cloud removal
generate adversarial network
group normalization
deep residual shrinkage network
loss function
