ABSTRACT

In a specific remote sensing study design, the utilization of images from a particular satellite is necessary. However, the images might be unavailable in a certain time range. Therefore, a conversion method from available remote sensing images at the time range is required. In this paper, we proposed machine learning models that are capable to convert Landsat-8 images to Sentinel-2 images. The models are inspired by the advancement of super-resolution model based on Deep learning. The result of this study shows that the proposed models can predict Sentinel-2 images which are quantitatively and qualitatively similar to the original images.

摘要

在特定的遥感研究设计中，必须利用来自特定卫星的图像。但是，图像可能在特定时间范围内不可用。因此，需要在该时间范围内从可用的遥感图像进行转换的方法。在本文中，我们提出了能够将Landsat-8图像转换为Sentinel-2图像的机器学习模型。这些模型的灵感来自基于深度学习的超分辨率模型的发展。这项研究的结果表明，所提出的模型可以预测Sentinel-2图像，该图像在数量和质量上都与原始图像相似。