Thanks to the last technological advances in the aerospace sector that have been occurring in the last decades, commercial companies have been able to launch affordable satellites that can capture regular high resolution images. On the other hand, thanks to the latest advances in GPU computation and data collection, deep learning has emerged as one of the most promising fields in artificial intelligence (AI). For instance, super-resolution (SR) of images using deep learning have topped the state of the art in the past years.
Comercial nano-satellites capturing high-resolution images have a shorter life due to their low Earth orbit (LEO), exposure to radiation, and most importantly the default components life. Altogether can affect the image quality.
One way of enhancing the image quality and therefore extending the satellite life can consist in using deep learning. Moreover, as we have witnessed with other types of images, SR can go beyond the pixel and yield a better and higher resolution. Although one risks hallucinating information, the chance of improving the image while helping objects and structures visibility via SR seems a promising research topic.
In this research project, we aim at exploring the benefits of using SR algorithms not just to increase the native resolution of the satellite images but also to enhance the image quality in terms of noise and blur. To accomplish this, a detailed search among current quality metrics and SR methods will be performed with the goal of developing new strategies and methods to increase the image quality.