SAR imaging from subsampled data: a composite compressed sensing/power spectrum estimation framework

Dr Gabriel Rilling; Research Fellow

23rd June, 2010 at 1.30 pm - Video Conference Room, Sanderson Building

Joint UoE/HWU videoconference seminar

Synthetic Aperture Radar (SAR) is an active ground imaging system basedon the processing of radar echoes acquired along the trajectory of anaircraft. For various reasons (compression, ability to temporarilyinterrupt data acquisition), there is a need for high quality imagingalgorithms using only a subset of the amount of data traditionallycollected by SAR acquisition platforms. This relates to the framework ofcompressed sensing which guarantees such a high quality reconstructionprovided the reconstructed image is compressible in some dictionary. Inthe case of SAR, only some parts of the images are compressible andcompressed sensing performs poorly on the remaining parts. Thismotivates the introduction of a composite framework where compressedsensing is used to reconstruct the compressible parts while anothertechnique is used for the remaining parts. Given the statistics of SARimages, power spectrum estimation techniques are a good candidate forthe non-compressible parts. We propose to extend such techniques to makethem applicable to subsampled SAR data and study their behaviour usingreal world data.