Harnessing Generative Adversarial Models to Enhance Satellite Imagery Quality

High-resolution satellite images are also very important for urban planning, environmental monitoring, and solar panel defect detection. Due to the high price and technical performance issues of conventional satellite imaging systems, however, low-resolution images are obtained mostly. This study investigates the use of Generative Adversarial Networks (GANs) for enhancing satellite image resolution, focusing on SuperResolution GAN (SRGAN) and Enhanced SRGAN (ESRGAN). On a dataset consisting of 400 images, the models were tested with perceptual and structural quality measures. Results show that ESRGAN best in terms of visual quality with smaller LPIPS values (0.08-0.14) but with slightly smaller PSNR (30.8033.24 dB) as compared to SRGAN, which has larger PSNR ($32.92-34.55 \mathrm{~dB}$) and coarser details. These results show that GAN-based methods constitute low-cost alternatives for improving the quality of satellite images, paving the way for large-scale applications in remote sensing, solar energy, and disaster relief.