Improving spatial resolution in satellite and aerial imagery is a fundamental issue in remote sensing that directly impacts the accuracy of data interpretation and application. Traditional satellite imagery, such as that obtained from Sentinel-2, often lacks the necessary spatial detail for precise analysis, limiting its effectiveness in applications like precision agriculture, environmental monitoring, and resource management. Enhancing the resolution of these images is essential for extracting finer details and improving the overall utility of the data, especially in real-world scenarios where high-quality imagery is crucial.

This project presents a series of advancements in super-resolution techniques specifically designed for remote sensing imagery. By integrating cutting-edge deep learning techniques and leveraging attention mechanisms, Channel-wise Gated Attention Network (CGA-Net) and the Real World Super-Resolution Generative Adversarial Network (RWSRGAN) have been proposed which aim to enhance the spatial resolution of low-resolution satellite images. This work utilizes real-world remote sensing data and addresses the complexities associated with non-ideal, noisy imagery through the application of resolution enhancement techniques of pan-sharpening or super-resolution. The results demonstrate significant improvements in image quality, contributing to more accurate analysis and broader applicability of satellite imagery in various remote sensing tasks.