부산대학교 도서관

In this paper, a deep learning-based framework for solving the inverse scattering (IS) problem in half-space medium is presented. First, the qualitative method of diffraction tomography (DT) is used to generate a low-resolution reconstruction image, using the scattered field data. Then, a multilayer real-valued convolutional neural network (CNN) is designed to enhance the resolution of the DT images. Using the initializing step, alleviates the learning procedure complexity and challenges. The preliminary reported results, reveal that the proposed network could obtain high quality reconstructions and performs better than conventional nonlinear inverse scattering methods in terms of both image quality and computational time.