基于Sentinel-2卫星遥感影像的去云方法研究

Research on Cloud Removal Method Based on Sentinel-2 Satellite Remote Sensing Image

在光学遥感卫星图像中,云是普遍存在的现象,它严重降低了图像的质量,因此,去云处理就是一个必不可少的步骤。深度神经网络在许多图像处理任务中取得了成功,但是利用该方法针对遥感图像的去云研究较少。本文采用GAN来解决遥感图像去云问题,首先训练生成模型生成无云影像,同时训练判别模型使生成的模型更加真实和清晰,最终达到从被云覆盖的卫星图像中恢复并增强这些区域的信息,生成质量更好的无云图像的目的。基于人工智能标注的Sentinel-2卫星遥感影像数据集的试验表明,与传统的小波变换基准相比,提出的生成对抗网络模型在去云处理方面效果有明显提升。

In optical remote sensing satellite images,cloud is a common phenomenon,it seriously reduces che quality of the image,so cloud processing is an essential step.Deep neural networks has been successful in many image processing tasks,but less research has been done on cloud removal methods for remote sensing images using this method.In this paper,we use GAN to solve the problem of remote sensing image cloud removal by first training the generative model to generate cloud-free images.At the same time,training the discriminative model to make the generated model more realistic and clear,and finally achieve the purpose of recovering and enhancing the information of these regions from the cloud-covered satellite images to generate better quality cloud-free images.Experiments based on the manually intelligent annotated Sentinel-2 satellite remote sensing image dataset show that the proposed generative adversarial network model is significantly more effective in cloud removal processing compared with the traditional wavelet transform benchmark.