Unrolled convolutional neural network for full-wave inverse scattering

Yarui Zhang (1) , Marc Lambert (1) , Aurélia Fraysse (2) , Dominique Lesselier (2)

1 GeePs - Laboratoire Génie électrique et électronique de Paris
2 L2S - Laboratoire des signaux et systèmes

An unrolled deep learning scheme for solving full- wave nonlinear inverse scattering problems (ISPs) is proposed. Inspired by the so-called unrolled method, an iterative neural network structure combining the contrast source inversion (CSI) method and residual network (ResNet) is designed. By embedding the CSI iterations into the deep learning model, the domain knowledge is well incorporated into the learning process. Thor- ough numerical tests are carried out to evaluate the performance, stability, robustness, and reliability of the proposed approach. Comparisons with the widely used U-net structure and CSI exhibit the advantage of the proposed approach.

Mots clés: contrast source inversion method; deep learning; inverse scattering; unrolled method

Domaines: Traitement du signal et de l'image [eess.SP]; Apprentissage [cs.LG]

Yarui Zhang, Marc Lambert, Aurélia Fraysse, Dominique Lesselier. Unrolled convolutional neural network for full-wave inverse scattering. IEEE Transactions on Antennas and Propagation, 2023, 71 (1), pp.947-956. ⟨10.1109/TAP.2022.3216999⟩. ⟨hal-03694636⟩