Context: SAR images can be acquired all time (day and night and whatever the weather) which is a real advantage for tropical areas. Nevertheless the signal is affected by speckle noise which induces strong fluctuations.
Project: The recent Noise2Noise framework  proposed to learn image restoration without clean data by feeding the network with two noisy images with the same conditional expected values. This framework is very useful when clean data are difficult to obtain which is the case for SAR imagery. This framework will be evaluated for SAR images in two situations. In the first one, the network is trained using simulated SAR data for input and target (the clean image being provided by multi-temporal processing). In the second situation, more realistic, two images taken at two different times are used to do the training. This situation implies an automatic checking to avoid using data with time variations. The code of  is available and the clean images (first case) and multitemporal stack (second case) are provided.
Supervision: Florence Tupin, Emanuele Dalsasso
 Noise2Noise: Learning Image Restoration without Clean Data,
Jaakko Lehtinen, Jacob Munkberg, Jon Hasselgren, Samuli Laine, Tero Karras, Miika Aittala, Timo Aila