IE-VAE: A Deep Learning Method for Solving Electromagnetic Inverse Scattering Problems Based on Variational Autoencoder

Nonlinear electromagnetic inverse scattering is a super-resolution imaging technique that, compared to traditional imaging, it can make full use of the more realistic interaction between scatterer and electromagnetic field, thus reconstructing the target image more accurately. However, traditional methods for solving inverse scattering problems face significant challenges due to their strong nonlinearity, ill conditioning, and expensive computational costs. To solve these difficulties, we propose a deep learning method called IE-VAE to solve the electromagnetic inverse scattering problem. The IE-VAE method consists of two parts: an encoder and a decoder. The encoder part is the process of solving the forward scattering problem with electromagnetism, and the decoder part is realized by an improved very deep convolutional networks network(VGG). The improved VGG network is designed for solving the strong nonlinearity and ill-posed problem of inverse scattering problems. The proposed IE-VAE method mainly consists of two steps: Firstly, cattering field data can be obtained by inputting the shape parameters and dielectric constant of the object into the encoder section to obtain. And then the improved VGG network acts as a decoder to obtain the corresponding dielectric constant graph in the domain of interest(DOI). Numerical experiments showed that the proposed network was computationally effective compared with other state-of-the-art methods, because the decoding processing steps are executed in real-time.