Without phase information of the measured field data, the phaseless data inverse scattering problems (PD-ISPs) counter more serious nonlinearity and ill-posedness compared to full data ISPs (FD-ISPs). In this paper, we propose a learning-based inversion approach in the frame of the U-net convolutional neural network (CNN) to quantitatively image unknown scatterers located in homogeneous background from amplitude-only measured total field (also denoted phaseless data). Three training schemes with different inputs to the U-net CNN are proposed and compared, i.e., the direct inversions (DIS) scheme with phaseless total field data, retrieval dominant induced currents by Levenberg-Marquardt (LM) method (PD-DICs) and phaseless data with contrast source inversion scheme (PD-CSI). We also demonstrate the setup of training data and compare the performance of the three schemes using both numerical and experimental tests. It is found that the proposed PD-CSI and PD-DICs perform better in terms of accuracy, generalization ability and robustness compared to DIS. PD-CSI has the strongest capability to tackle with PD-ISPs, which outperforms the PDDICs and DIS.

The electromagnetic inverse scattering problem can be effectively handled by means of the so-called distorted-Born iterative method (DBIM). A new method, denoted as the subspace-based DBIM (S-DBIM), is proposed. It updates the Green’s function for the inhomogeneous background at each step of the iterative procedure, like DBIM. By linearly retrieving the deterministic subspace of the induced current, the S-DBIM estimates the total electric field more accurately than the DBIM does and thus exhibits a faster convergence speed. To avoid the computationally demanding and arguably hard choice of the optimal Tikhonov regularization term involved in the procedure, a second version of S-DBIM, which is more robust against noise and has faster convergence, is also proposed. Thorough numerical simulations show that both versions of S-DBIM achieve super-resolved retrievals of the benchmark “Austria” profile.

Deep learning based inverse scattering (DL-IS) methods attract much attention in recent years due to advantages of fast speed and high-quality reconstruction. The loss functions of neural networks in DL-IS methods are commonly based on a pixel-wise mean squared error (MSE) between the reconstructed image and its reference one. In this paper, we introduce a structural similarity (SSIM) loss function to combine with the MSE loss for reconstructing dielectric targets under a DL-IS framework. The SSIM loss imposes a further regularization on the target at the perceptual level. Numerical tests for both synthetic and experimental data verify that this new perceptually-inspired loss function can effectively improve the imaging quality and the generalization capability of the trained model.

A deep learning-based inversion algorithm is developed to solve the inhomogeneous background inverse scattering problem (ISP). To alleviate the burden of nonlinearity and ill-posedness of the ISP, a noniterative method called the distorted-Born backpropagation scheme is introduced to quantitatively reconstruct a rough image of the unknown object in inhomogeneous background. The roughly reconstructed result serves as the input of the designed generative adversarial network (GAN), which outputs the fine reconstructed image of the relative permittivity. The generator network of the GAN is well designed as an encoder-decoder structure configured with the attention scheme. The discriminator network is taken to supervise the generator to learn the features of target scatterers through an adversarial training process. The proposed method is proven to be effective in reconstructing objects embedded in inhomogeneous background, which promises a real-time application future of the ISP.

In this communication, a rectangular dielectric resonator antenna (DRA) is integrated with a backed cavity to enlarge the DRA side length by 1.8 times. The backed cavity also reduces the sensitivity of the DRA resonance to the DRA size, resulting in a relaxed fabrication tolerance. Based on the enlarged DRA design, a stacked DRA (sDRA) is proposed to improve the bandwidth from 8.7% to 15% and enhance the realized gain by an average of 1.5 dB. The proposed mechanisms are verified by fabricating a 4x4 sDRA array with an enlarged DRA side length of 1.8 mm (0.40λ), an improved bandwidth of 62.7-73.9 GHz (16.4%) and an enhanced measured gain of up to 17.2 dBi. The enlarged DRA dimension permits a higher DRA operating frequency without reducing the DRA size. The proposed millimeter-wave (mmW) sDRA array provides a wide bandwidth and high-gain solution for 67 GHz unlicensed band communications.

In this work, we introduce a learning-based method to achieve high-quality reconstructions for inverse scattering problems (ISPs). Particularly, the proposed method decouples the full-wave reconstruction model into two steps, including coarse imaging of dielectric profiles by the back-propagation scheme, and a resolution enhancement of coarse results as an image-to-image translation task solved by a novel perceptual generative adversarial network (PGAN). A perceptual adversarial (PA) loss, which is defined as a perceptual loss for the generator network using hidden layers from the discriminator network, is employed as a structural regularization in PGAN. The PA loss is further combined with the pixel-wise loss, and also possibly the adversarial loss, to enforce a multi-level match between the reconstructed image and its reference one. The adversarial training of the generator and discriminator networks ensures that the structural features of targets are dynamically learned by the generator. Numerical tests on both synthetic and experimental data verify that the proposed method is highly efficient and it achieves superior imaging results compared to other data-driven methods. The validation of the proposed PGAN on ISPs also provides a fast and high-precision way for solving other physics-related imaging problems.