15 January 2021 in IEEE Antennas and Propagation Magazine
Biomedical imaging is a relevant noninvasive technique aimed at generating an image of the biological structure under analysis. The arising visual representation of the characteristics of the object is affected by both the measurement process and reconstruction algorithm. This procedure can be considered as a hybridization of data information, measurement physics, and prior information.
9 August 2021 in IEEE Transactions on Antennas and Propagation
In this work, we design an iterative deep neural network to solve full-wave inverse scattering problems (ISPs) in the 2-D case. Forward modeling neural networks that predict the scattered field are embedded in an inversion neural network. In an iterative manner, the inversion network predicts the model update from the residual between the simulated data and the observed data. The proposed inversion network can achieve super-resolution reconstruction meanwhile keeping the simulated data of reconstructed models well consistent with the observed data. We validate this method with both synthetic and experimental data inversion. Results show that the inversion network can predict models with high accuracy, efficiency, and good generalization ability. By combining deep learning and physical simulation together, the proposed method provides a way for real-time imaging with high reliability and accuracy.
1 November 2022 in IEEE Transactions on Antennas and Propagation
In this paper, we propose the neural Born iterative method (NeuralBIM) for solving 2D inverse scattering problems (ISPs) by drawing on the scheme of physics-informed supervised residual learning (PhiSRL) to emulate the computing process of the traditional Born iterative method (TBIM). NeuralBIM employs independent convolutional neural networks (CNNs) to learn the alternate update rules of two different candidate solutions regarding the residuals. Two different schemes are presented in this paper, including the supervised and unsupervised learning schemes. With the data set generated by the method of moments (MoM), supervised NeuralBIM are trained with the knowledge of total fields and contrasts. Unsupervised NeuralBIM is guided by the physics-embedded objective function founding on the governing equations of ISPs, which results in no requirement of total fields and contrasts for training. Numerical and experimental results further validate the efficacy of NeuralBIM.
30 July 2021 in IEEE Transactions on Microwave Theory and Techniques
In this work, we present a deep-learning-based low-frequency (LF) data prediction scheme to solve the highly nonlinear inverse scattering problem (ISP) with strong scatterers. The nonlinearity of ISP is alleviated by introducing the LF components in full-wave inversion. In this scheme, a deep neural network (DNN) is trained to predict the absent LF scattered field data from the measured high-frequency (HF) data. Then, a frequency-hopping technique is applied to invert the predicted LF data and measured HF data, where the inverted LF model is served as an initial guess for the HF data inversion. In this way, the risk of HF data inversion getting trapped in local minima is largely reduced. Furthermore, an iterative training method is employed to continuously update the DNN based on the previous inverted model to predict more accurate LF data, thereby improving the reconstruction result. Both synthetic and experimental results are performed to verify the efficacy of this scheme.
8 August 2022 in IEEE Transactions on Antennas and Propagation
We present an inversion algorithm with a deep-learning-based model compression scheme. Models are described with latent parameters of a trained variational autoencoder (VAE) neural network. Given observed data, latent parameters are inverted by minimizing the data misfit cost function using the Gauss-Newton method. This inversion algorithm is tested using both synthetic and experimental datasets. We achieve a 0.87% compression rate while maintaining high-quality reconstruction. The deep neural network renders nonlinear model compression, which largely reduces the number of unknowns; hence, it has higher computational efficiency. Furthermore, various prior knowledge that is difficult to describe with rigorous forms can be incorporated into inversion through training the neural network, which mitigates the ill-posedness of the inverse problem.
2 February 2021 in IEEE Antennas and Wireless Propagation Letters
Solving the combined field integral equation (CFIE) for the large-scale scattering problem is computationally expensive. In this letter, we investigate the feasibility of applying deep learning to solve the CFIE for 2-D perfect electrically conducting objects. Inspired by the conjugate gradient method, an iterative deep neural network is designed to learn the manner of solving the surface current density from the CFIE, with the input being the coefficient matrix of the equation. This process involves physics through surface integration and need less iterations than the conventional iterative equation solver. In numerical tests, we evaluate the network's performance by comparing the predicted surface current density and bistatic scattering cross section with the solutions rigorously computed. This method provides an insight into applying machine learning techniques together with electromagnetic (EM) physics to fast EM computation with the same level of accuracy as traditional method.