A two-stage algorithm based on convolution and attention hybrid architecture for multi-obstacle inverse scattering problems: Integrating obstacle perception and shape recognition

Abstract In this paper, we consider the use of an artificial neural network approach to solve the multi-obstacle inverse scattering problems. We propose a two-step algorithm consisting of obstacle perception and shape recognition. In the obstacle perception step, we train a deep neural network to predict the number of obstacles in the system of interest. At the shape recognition step, we train multiple neural networks to invert the shape of each obstacle. A large number of numerical experiments are provided to verify the effectiveness of the proposed method. It is worth pointing out that our algorithm is not only applicable to the full-phase data problem, but also to the phaseless data problem, and even to the finite aperture problem.