Enhancing Satellite Image with Enhanced Super Resolution GAN

Satellite imagery finds wide-ranging applications across various disciplines, ranging from environmental monitoring to urban planning and disaster management. Owing to sensor resolution limitations and atmospheric noise, satellite images are generally low-resolution and visually poor. In this paper, we discuss the use of Enhanced Super-Resolution Generative Adversarial Networks to super-resolve and improve satellite images. ESRGAN, a novel deep learning model, uses residual-in-residual dense blocks and perceptual loss functions to produce high-fidelity images. Through our experiments, we show that ESRGAN performs much better than traditional interpolation methods and baseline super-resolution models in terms of PSNR and SSIM metrics. The algorithm proposed here presents a useful tool for the enhancement of satellite imagery, enabling enhanced decision-making in remote sensing.