Compression of synthetic-aperture radar images

Abstract

In this paper the problem of the synthetic-aperture radar images compression is considered. The algorithm of canonical coherent scatterers identification, proposed in [1, 2], based on the analysis of polarimetric signatures, is the starting point of the studies. The question whether the significant dimension reduction of the SAR image matrix preserves the information encoded in the SAR picture or not, is the topic of the paper. It turns out that the compression, by using the Kohonen neural network, allows us to reduce the dimension of the data from 16200-component vector to 100-component vector without losing information. The studies are led in the context of polarimetric data that encode full information about the scatterer. However, there are essential problems with such data processing. Therefore the topic is crucial in the context of the SAR images analysis.