Abstract

Generative adversarial neural networks are used to increase the resolution of satellite images of a certain class without additional data. The quality of the obtained high-resolution images is assessed by the signal-to-noise ratio and the measure of structural similarity. Based on the known loss functions used in generative adversarial neural networks, a function specific to the problem being solved is obtained. Training and testing is carried out on the example of images of a railway infrastructure, covering about 78 km of railways.

中文翻译：

---

基于生成神经网络的全色卫星图像空间分辨率提高

摘要

生成对抗性神经网络用于提高特定类别卫星图像的分辨率，而无需其他数据。通过信噪比和结构相似性度量来评估获得的高分辨率图像的质量。基于在生成对抗神经网络中使用的已知损失函数，可以获得特定于要解决的问题的函数。培训和测试以铁路基础设施的图像为例，覆盖了大约78公里的铁路。