Super-Resolution of GK2A/AMI Level-1B Imagery Using AI Models

Super-resolution is a technique for enhancing the spatial resolution of images, and a wide range of models has been actively proposed in recent years with advances in artificial intelligence (AI).This study aims to develop a simulated imagery generation technique that accommodates the enhanced spatialresolution requirements (from 2 km to 1 km and from 0.5 km to 0.25 km) of the GEO-KOMPSAT-5 (GK5) mission, scheduled for launch in 2031.To this end, we performed super-resolution on GK2A Level-1B imagery.The super-resolution methods used in this study are (1) a convolutional neural network (CNN), (2) a lightweight hybrid model integrating CNN and transformer components (HNCT), and (3) a traditional interpolation method, bicubic interpolation.Performance was evaluated both qualitatively and quantitatively using super-resolved outputs generated from low-resolution inputs and from the original images.The results show that, during daytime, the CNN produced the sharpest super-resolved images, achieving a peak signal-to-noise ratio (PSNR) above 28.7 dB, a structural similarity index (SSIM) above 0.89, and a root mean square error (RMSE) of approximately 1.8 K.At nighttime, however, the CNN yielded blurrier results, while bicubic and HNCT showed higher accuracy.Future work will focus on enhancing model performance through architectural refinement and retraining to meet the target resolution requirements of GK5.