Application of Generative Adversarial Network for Electromagnetic Imaging in Half-Space

We developed a novel deep-learning technique for reconstructing a conductor incident polarization wave. Due to the constraints of measurements, there are limitations on the measurement angle of the polarization wave. The measured scattered fields are input into the Direct Sample Method (DSM) to generate an initial guess image. This process effectively mitigates the highly nonlinear phenomena. Next, the DSM images are input into the Neural Networks. Numerical results showed that CNN had better performance than GAN in reconstructing electromagnetic imaging.