Neural network-assisted reconstruction of full polarimetric SAR information

This paper describes a novel approach to the reconstruction of synthetic aperture radar (SAR) fully polarimetric data from compact polarimetry (CP) ¿/4 mode. A method is developed which utilises a multi-layer perceptron (MLP) based neural network, to perform reconstruction of scenes with various ground-cover types. In particular, the approach shows potential for the reconstruction of full polarimetry for built-up areas as a complement to existing techniques which are more suitable for natural land cover areas. Performance assessment is presented, using both L-band and C-band data, involving comparison with existing techniques using mean-squared and mean-squared-log measures.