Central AsianResearch Index

The use of generative models (gan) to improve the quality of satellite images

This article discusses the use of generative competitive networks (GANS) to improve the quality of satellite images. Satellite imagery plays a key role in environmental monitoring, mapping, and territory analysis. However, data quality often deteriorates due to low resolution, atmospheric interference, and hardware limitations. Generative models allow solving these problems by super-resolving images, removing noise, and reconstructing lost fragments. The article analyzes the main methods of using GAN, such as SRGAN and ESRGAN to increase resolution, Pix2Pix and CycleGAN to convert images between different spectral ranges. Examples of successful application of these technologies in geoinformation systems, environmental monitoring and remote sensing of the Earth are considered. Key challenges such as the complexity of GAN training, the need for large amounts of training data, and computing resources are also discussed. Nevertheless, the further development of neural network architectures and the integration of GAN with other machine learning methods open up new prospects for improving the quality of satellite data. The use of generative models in satellite image processing contributes to more accurate data analysis and improves their interpretation capabilities.

Сведения

Опубликовано
2025 · Т. 98, № 4, С. 64-72
Язык
Английский
ISSN
2709-4693 · 2709-4707
Издатель
National Engineering Academy of the Republic of Kazakhstan