Deep Neural Network for Super-resolution of Multitemporal Remote Sensing Images

Book excerpt: Since few years ago, artificial intelligence (AI) has become a spotlight technology in which a lot of people are interested in. Most of them want to do research and use it to solve a huge variety of modern and difficult computing problems which could be associated with a wide variety of interesting fields. As soon as AI has improved, convolutional neural networks (CNN) have taken an excellent role in the world of image processing, in particular, for Remote Sensing applications. Nevertheless, artificial intelligence for multi-image superresolution from multi-temporal imagery has received little attention so far. In this work, it is proposed a CNN, which exploits both spatial and temporal correlations in the low-resolution images by using two different convolutional layers (2D and 3D convolutions) to combine multiple satellite images from the same scene which are taken in different temporal moments. The experiments have been carried out using a dataset generated by Sentinel-2 (European Space Agency satellite) images captured over 2 different places over the world, New York and El Cairo. This model aims to obtain super-resolution images from five low-resolution images, or less, being aware of the number of input images that the CNN has.