Quantitative Imaging by Generative Adversarial Network with Data Complementation for Limited-aperture Inverse Scattering Problem

Practical applications of the quantitative inversion are restricted by the limited availability of measurement data, which is denoted as the limited-aperture inverse scattering problem (ISPs). To circumstance this challenge, a novel limitedaperture data complementation method utilizing generative adversarial network (GAN) for quantitative imaging is presented for the first time. Based on the powerful nonlinear fitting capabilities and low computational complexity, the pix2pix GAN is utilized to complement the measured limited-aperture data into comprehensive full-aperture data at the first step. Afterward, the procedure involves the utilization of a combination of a basic backpropagation (BP) algorithm and another pre-trained pix2pix GAN to reconstruct the permittivity image of scatterers from the recovered full-aperture data. The proposed method has the potential to address the challenge that existing methods encounter in reconstructing scatterers under limited-aperture, particularly at relatively small aperture, such as 180°. Simulations reveal that the proposed method effectively reconstructs permittivity images, achieving impressive results with the mean square error (MSE) less than 0.0999 and structural similarity index (SSIM) over 86.03% in MNIST dataset. Strong robustness and generalization performance are also demonstrated on other test datasets.