Super Resolution of Resourcesat-2 Satellite Imagery using SRCNN

Remote sensing technologies play a pivotal role in monitoring and understanding our changing environment. However, acquiring high-resolution satellite imagery poses challenges, often limited by cost and availability. This research addresses the critical need for enhanced spatial resolution in satellite imagery through the application of Convolutional Neural Networks (CNNs)[1]. Specifically, we focus on the super-resolution of Resourcesat-2 LISS-3 imagery, a widely utilized remote sensing dataset. Given the 24-meter resolution of the original imagery, obtaining higher-resolution (5-meter) data is a daunting task. In response to this challenge, our proposed method leverages CNNs to merge the 24-meter data into 5-meter data, thereby enhancing the overall resolution. This approach not only circumvents the limitations associated with obtaining high-resolution data but also contributes to the broader goal of improving the interpretability and applicability of satellite imagery in diverse fields. Our experimental results demonstrate the efficacy of the proposed method, showcasing the potential for cost-effective and efficient enhancement of spatial resolution of remote sensing data. The findings presented herein contribute to advancing the capabilities of satellite-based observation systems, opening avenues for improved visualisation, monitoring and analysis of Earth’s dynamic landscapes.