A novel adaptive wide-angle SAR imaging algorithm based on Boltzmann machine model

摘要

Scattering dependency often exists in both the spatial location and the viewing angle. Based on the assumption of isotropic point scattering model, however, conventional narrow-angle synthetic aperture radar (SAR) imaging algorithms have been no longer suitable to the scattering dependency model. To improve azimuth resolution and capture richer observation information, sparsity-driven (SD) wide-angle SAR (WSAR) imaging algorithms have been developed. Actually, existing SD-based WSAR imaging algorithms are sensitive to the regularization parameters which are required to adjust manually. These methods indeed limit their practical applications. To solve this problem, in this paper, we propose an adaptive WSAR imaging algorithm based on the Boltzmann machine (BM) model. In particular, we model the spatial sparsity and high azimuth correlation of scattering energy by virtual of a special BM prior. Then, the support of sparse representation and imaging parameters including BM parameters, noise variance and the variance of each sparse representation element are jointly estimated by a block-coordinate descent process. Finally, the proposed WSAR imaging algorithm is performed adaptively via sparse representation. Experiments are conducted by synthetic scene and simple tank dataset of high-frequency electromagnetic scattering calculation software. Extensive empirical results demonstrate that the proposed algorithm can achieve better imaging performance than the conventional algorithms in terms of relative mean squared error and support identification error.