Convolutional RNN Based on Multi-Scale Features for Solving Inverse Scattering Problems

In this paper, a novel deep learning-based inversion method with a multi-scale structure is developed for solving electromagnetic inverse scattering problems (ISPs). The basic idea of proposed method is to enable the neural network to reconstruct the relative permittivity profile from coarse to fine. By learning multi-scale features, the cascaded convolutional recurrent neural network (CRNN) can capture more details of objects and promote finer reconstructed images. The effectiveness of the proposed method is verified by some numerical tests and the imaging quality is improved in comparison to the traditional U-Net.