An Enhanced U-Net Scheme with Self-Attention Mechanism for Inverse Scattering Problems

The nonlinearity and ill-posedness of the inverse scattering problems make them difficult to solve. Multiple scattering effects are the main cause of nonlinearity. In this paper, a deep learning method with self-attention mechanism is proposed to address the difficulties caused by multiple scattering effects. The main block of the proposed network architecture combines self-attention layers and convolutional layers, enhancing the global perception ability of the network. Numerical results show that the proposed method achieves higher accuracy in solving the inverse scattering problem compared to architectures relying solely on convolutional neural networks.