Deep Learning based fusion of SAR and optical data to synthesize cloud-free and enhanced quality imagery

The presence of noises, clouds and another atmospheric phenomena represent an obstacle for optical remote sensing applications. A commonly used strategy is the composition of images acquired in several different dates into a single cloud-free imagery. This approach requires, in most cases, generates several unexpected artifacts and also requires a good and accurate cloud mask which, however, is not always available. Active sensors, due to the larger wavelengths used, are able to overcome the clouds and produce images at any atmospheric conditions. Despite that potential, those images are usually more complex and harder to be interpreted. Several researches have demonstrated the potential of Deep Learning methods, especially generative networks, to fuse SAR and optical imagery, synthesizing cloud free and enhanced quality images. In this sense, this MSc topic focus on designing and implementing efficient Deep Learning methods to perform this task.