Complex-Valued Pix2pix—Deep Neural Network for Nonlinear Electromagnetic Inverse Scattering

Nonlinear electromagnetic inverse scattering is an imaging technique with quantitative reconstruction and high resolution. Compared conventional tomography, it takes into account the more realistic interaction between internal structure of scene waves. However, there are still open issues challenges due to its inherent strong non-linearity, ill-posedness computational cost. To overcome these shortcomings, we apply image translation network, named as Complex-Valued Pix2pix, on problem field. Pix2pix includes two parts Generator Discriminator. The employs a multi-layer complex valued convolutional neural while Discriminator computes maximum likelihoods original value reconstructed from aspects complex: real part imaginary part, respectively. results show that can learn mapping initial contrast in microwave models. Moreover, introduction discriminator, capture features nonlinearity than traditional Convolutional Neural Network (CNN) by confrontation training. Therefore, without considering time cost training, may be effective way solve problems other deep learning methods. main improvement this work lies realization Generative Adversarial (GAN) problem, adding discriminator method optimize network It has prospect outperforming methods terms both quality efficiency.