Speckle Suppression Algorithm Of PolSAR Image Based On Sparse Representation

Polarimetric Synthetic Aperture Radar(PolSAR) can transmite and receive different polarimetric electromagnetic wave in the same time, so it can access more polarimetric information of targets than single-polarimetric SAR. At present, it has been widely used in geographic surveying and mapping, agroforestry, urban planning, disaster prevention and mitigation, marine science, military science and other fields. However, speckle noise exist in the SAR images and affect further application of SAR images. Therefore, speckle reduction is a key step in PolSAR image preprocessing.In this paper, speckle characteristics and sparse representation theory are firstly introduced, and then two effective sparse filtering algorithm based on PolSAR are proposed to solve the speckle noise. One is a speckle reduction algorithm based on double-scale dictionary of polarimetric SAR image. first, original image is divided into homogeneous or heterogeneous area according to the homogeneous discriminant criterion. then, large-scale dictionary is used in homogeneous area, small-scale dictionary is used in heterogeneous areas,KSVD algorithm and OMP algorithm are used to dictionary training and sparse denoising respectively. finally, the denoised results are reconstructed and synthetized, forming the final denoised SAR image. Experimental results with AIRSAR data show that, the proposed algorithm's speckle noise suppression ability is stronger, the edge texture details is more clearly compared with the results of the single-scale dictionary sparse denoising algorithm, in addition, the strong scattering point target amplitude signature preservation and polarization characteristics preservation of SAR image can obtain a good effect.The second method is multi-polarimetric SAR image joint sparse filtering algorithm.firstly, similar structer information among different polarimetric SAR channel are used to construct joint sparse model. then, joint sparse coefficient of SAR image is solved by this model. finally, clean images of each SAR channel are reconstructed. Experimental results with the same data show that the proposed method is effective both on speckle reduction and strong scattering point target signature preservation compared with the results of each channel image sparse denoising separately.