SAR Phase Unwrapping with Hopfield Neural Network.

Interferometric images obtained with synthetic aperture radar (SAR) processing are the starting point for high resolution digital elevation maps (DEM). Bi-dimensional phase unwrapping is the key point for the correct generation of those maps. This problem may be solved with a deterministic approach only for ideal conditions. On the contrary, when noise, volume and casual phase effects or when abrupt changes in terrain topology or low backscatters are taken into account, phase inconsistencies appear in the unwrapping process (ghost lines) and lead to incorrect phase reconstruction. Here it is presented an algorithm for automatic phase unwrapping based on a Hopfield neural network approach that may be employed in real case conditions. Several tests have been carried out on both synthetic and real images and the result are presented along with future algorithm enhancements.