Skip to Main Content
Synthetic aperture radar (SAR) is a sophisticated remote sensing tool that is capable of providing high resolution images from a moving platform. Due to the very poor correlation and high entropy of SAR raw data, redundancy reduction techniques have not proven successful and a lossy compression is necessary. In a previous work, we have presented a compression of the raw SAR signal using five kinds of wavelets. The quality reconstruction was very good, however, due to noise like characteristics of the raw SAR signal, none of the standard wavelets was very efficient in compacting energy in the transform domain. In this paper, we propose to determine an optimal 2-D wavelet which is learned directly from the raw SAR data. The optimality criterion in the learning processes is redundancy minimization in the transform domain. Experiments show that this optimal wavelet performs better than the standard wavelets.