A Review and Analysis of GAN-Based Super-Resolution Approaches for INSAT 3D/3DR Satellite Imagery using Artificial Intelligence

Abstract

The Indian National Satellite System (INSAT)-3D/3DR is a geostationary satellite that is used for meteorological applications in the Indian region. Geostationary satellites have significant spatial coverage and good temporal resolution that help to monitor the evolution and propagation of meteorological systems. Meteorologists use satellite images to observe the locations of severe weather and understand the physical processes involved in the system. Image Super-Resolution (SR) aims to convert low-resolution images into high-resolution images while maintaining image quality. The SR techniques will improve the visualization of convective systems and tropical cyclones, facilitating accurate location-based warnings. This paper presents a comparative comparison of computer models for converting Low-Resolution (LR)(INSAT)-3D/3DR images into super-resolution images. This study also discusses and investigates the various Generative Adversarial Network (GAN)-based models, including the Super Resolution Generative Adversarial Network (SRGAN), Enhanced Super Resolution Generative Adversarial Network (ESRGAN), and Real Enhanced Super Resolution Generative Adversarial Network (Real-ESRGAN). The findings are compared to established approaches such as Bicubic Interpolation and Super- Resolution Convolution Neural Network (SRCNN). This study demonstrates that Real-ESRGAN performs better on weather satellite images than other cutting-edge approaches.