Deep Learning For Super-Resolution Of Unregistered Multi-Temporal Satellite Images

Recently, convolutional neural networks (CNN) have been successfully applied to many remote sensing tasks. However, deep learning for multi-image superresolution from multitemporal imagery has received little attention so far. We propose a residual CNN that exploits both spatial and temporal correlations in the low-resolution image set by using 3D convolutional layers to combine multiple images from the same scene. The experiments have been carried out using a dataset of PROBA-V satellite ground images, composed of several low-resolution and high-resolution images taken at different times from instruments on the same platform, in the context of a challenge issued by the European Space Agency.