Abstract:Nonlinear electromagnetic (EM) inverse scattering is a quantitative and super-resolution imaging technique, in which the more realistic interactions between the internal structure of scene and EM wavefield has been taken into account the imaging procedure, in contrast to conventional tomography. However, it poses important challenges arising from its intrinsic strong nonlinearity, ill-posedness and expensive computation costs. In this work, we build the connection between the CNN and the unfold iterative solution to nonlinear electromagnetic inverse scattering, and then establish a novel CNN, termed as DeepNIS, for the nonlinear electromagnetic inverse scattering. The DeepNIS is a non-iterative solution to the nonlinear electromagnetic inverse scattering. A central issue to the DeepNIS-based solution is the convolution operation, which can be implemented in parallel. The noniterative and parallelizable natures of the DeepNIS imaging mechanism make it very suitable for dealing with large-scale inverse scattering problems. We showed that the DeepNIS has the overwhelming advantage over conventional inverse scattering methods in terms of the image quality and computational time. Our experimental results demonstrate that the proposed DeepNIS are capable of learning the inverse mapping between the raw intensity image and object directly from the experimental data. The results also suggest that the neural network can “learn” the governing equations of the system, including its forward operator and possible deviations from underlying idealizations and assumptions. Nonetheless, we still remark that one can explore more advanced CNN architectures to get more impressive inverse scattering results. To summarize, the DeepNIS could remarkably break the limitation of the conventional inverse scattering strategies, and can be used for treating the nonlinear electromagnetic inverse scattering with large-scale and high-contrast objects.

Physical model simulator-trained neural network for computational 3D phase imaging of multiple-scattering samples

Solving the Inverse Scattering Problem in Reflection-Mode Dynamic Speckle-Field Phase Microscopy (conference Presentation)

Wavefront Reconstruction Based on Deep Transfer Learning for Microscopy.

Universal and High-Fidelity Resolution Extending for Fluorescence Microscopy Using a Single-Training Physics-Informed Sparse Neural Network

Robust single-shot 3D fluorescence imaging in scattering media with a simulator-trained neural network

Real-time 3D Nanoscale Coherent Imaging via Physics-aware Deep Learning

Physics-informed Deep Neural Network Reconstruction Framework for Propagation-Based X Ray Phase-Contrast Computed Tomography with Sparse-View Projections

Intensity and phase imaging through scattering media via deep despeckle complex neural networks

Recognizing three-dimensional phase images with deep learning

AutoPhaseNN: unsupervised physics-aware deep learning of 3D nanoscale Bragg coherent diffraction imaging

A Novel Electromagnetic Reconstruction Algorithm for Dielectric Objects Using Neural Networks

Imaging reconstruction through long-range scattering media by using deep learning

Three-dimensional coherent X-ray diffraction imaging via deep convolutional neural networks

Physics-Informed Ellipsoidal Coordinate Encoding Implicit Neural Representation for high-resolution volumetric wide-field microscopy

Physics-informed neural networks for diffraction tomography

Phase Imaging through Scattering Media Using Incoherent Light Source

AI-enabled high-resolution scanning coherent diffraction imaging

Refractive index tomography with a physics based optical neural network

Reconstructing 3D Biomedical Architectural Order at Multiple Spatial Scales with Multimodal Stack Input

Nonlinear Electromagnetic Inverse Scattering Using DeepNIS