Inhomogeneous Media Inverse Scattering Problem Assisted by Swin Transformer Network

A deep learning-assisted inversion method is proposed to solve the inhomogeneous background imaging problem. First, a noniterative method called the distorted-Born modified Born approximation (DB-MBA) method is introduced, which retains a major part of the multiple scattering information of the unknown scatterers without resourcing to the time-consuming iteration. DB-MBA offers better reconstruction accuracy for unknown objects embedded in inhomogeneous media, compared to the traditional noniterative methods such as backpropagation scheme (BPS) and Born approximation (BA) method that disregard the multiple scattering effect. To further retrieve the remaining part of multiple scattering fields that accounts for the super-resolution information, the result obtained by DB-MBA serves as the input to a well-trained Swin Transformer network. The attention mechanism involved in shifted window enables the algorithm to capture the global interactions between the objects, thus improving the performance of the inhomogeneous background imaging and at the same time reducing the computational complexity. The effectiveness of the proposed method is demonstrated using both synthetic data and experimental data. Super-resolution imaging is achieved with real-time speed, indicating the fast and high reconstruction ability of the proposed method.