The clarity and the utility of remote sensing data is improved by super resolution of satellite images. This research-oriented study uses deep learning models which include SRCNN, LAPSRN, RCAN, SRGAN, Fine-tuned ESRGAN and A-ESRGAN with U-Net Discriminator for super resolution. Open source DOTA dataset was used in which the high resolution (1024 X 1024) patches were downscaled to low resolution (256 X 256) patches to simulate low resolution image scenarios of real world. Both type of images was used to fine tune the real ESRGAN model. The fine-tuned ESRGAN performed best on the given test dataset, with a maximum PSNR of 44.39, mean PSNR of 30.85, and median PSNR of 28.11. The A-ESRGAN model, which used a single channel U-Net discriminator, achieved a maximum PSNR of 41.68 and a mean PSNR of 26.13. The traditional deep learning models which were used for super-resolution tasks like SRCNN and LAPSRN performed significantly lower than the modern architectures, with SRCNN achieving a mean PSNR of 24.04. Later the visual improvements in the high and low-resolution images were compared by passing them through a finetuned Yolov8 model, where the Yolov8 model was able to detect more artifacts and objects on the high-resolution images than the low-resolution images. The outcome of our study demonstrates the significance of using modern deep learning models in combination with domain-specific training and fine-tuning. These outcomes are important in domains that require better satellite clarity, with future work aimed at optimizing these models for real-time use on edge and IOT devices.