Compression of SAR interferograms for parameter retrieval using neural networks

SAR interferograms are generally affected by different types of errors. Phase noise in interferometry is introduced by the radar system, by the propagation path through the variably refractive atmosphere, by spatial decorrelation of the electromagnetic fields scattered back from the surface elements. In many applicative cases, such as DEM generation, a pixel based information is required and noise can be reduced using a multilook technique which is often applied by averaging neighboring pixels. In other cases, the pixel based information is less important with respect to the fringes distribution pattern observed over the area of interest. More specifically, in applications regarding tectonics, the retrieval problem is often focused on the estimation of the fault parameters from the InSAR differential interferogram whereas this latter is generated by computing the phase difference of two radar images, acquired before and after an earthquake, on a pixel-by-pixel basis. Elements such as the shape and periodicity of the fringes, the number of lobes and their orientation represent the information contained in the interferogram. In such a case, besides the noise mitigation, it is also important to express the relevant information after having applied to the image some feature extraction technique, in order to avoid to design inversion algorithms receiving as input the value of each single pixel. The issue can be addressed by means of a spatial sampling, but this is not certainly an optimum solution for the problem. As far as we know, no specific techniques for dimensionality reduction applied to SAR interferograms have been presented in literature. In this paper two standard image filtering approaches based on harmonic analysis and a novel one based on autoassociative neural networks (AANN) are analysed. A specific application for the estimation of tectonic parameters from SAR interferometry is also presented.