Model-Based Polarimetric SAR Decomposition: An L1 Regularization Approach

Model-based polarimetric decompositions are often used to generate scene classifications from polarimetric synthetic aperture radar (POLSAR) imagery. The classification quality largely depends on whether the adopted polarimetric scattering mechanisms match the major in-scene scattering mechanisms. However, in-scene scattering variations still cause potential mismatches, resulting in unphysical or inaccurate decomposition results. The robustness of model-based polarimetric decompositions remains a general concern. In this article, we address the robustness of model-based polarimetric decomposition to variations of the in-scene scattering mechanisms using simulated datasets. The known simulated scattering mechanisms provide the needed “ground truth” against which we quantitatively evaluate the robustness of polarimetric decompositions. Accurate retrieval of known scattering mechanisms from simulated polarimetric data led us to develop a robust model-based decomposition. We propose a new approach to solve model-based decomposition by employing an <inline-formula> <tex-math notation="LaTeX">$L_{1}$ </tex-math></inline-formula>-regularized optimization procedure, which automatically selects a set of optimal polarimetric scattering mechanisms and guarantees nonnegative powers for the selected scattering mechanisms. We illustrate this <inline-formula> <tex-math notation="LaTeX">$L_{1}$ </tex-math></inline-formula> model decomposition by employing both simulated datasets and actual POLSAR imagery. Our new <inline-formula> <tex-math notation="LaTeX">$L_{1}$ </tex-math></inline-formula>-regularized approach to POLSAR model-based decomposition that mitigates observed biases seen in earlier decompositions provides robust scattering mechanism estimates and eliminates unphysical, negative scattering powers.