LAUSR

This article presents an efficient and accurate 3-D quantitative hybrid microwave imaging (MWI) method. The linear sampling method (LSM) is first carried out to quickly find the approximate shapes and locations of the unknown objects in the imaging domain based on the scattered field data recorded by receivers which are placed in the far-field zone and wrap the domain. Then the full-wave inversion (FWI) is implemented in a downsized domain which tightly encloses the unknown objects instead of in the whole domain through the Born iterative method (BIM) to quantitatively retrieve the dielectric model parameters of the objects. Because the LSM fails to obtain the sufficiently accurate shapes of the unknown objects, a trained 3-D convolutional neural network (CNN) U-Net is inserted between the LSM imager and the BIM solver to further refine the obtained shapes of LSM, which is expected to aid the following FWI. The proposed hybrid method is validated via the quantitative imaging of both inhomogeneous isotropic scatterers and multiple homogeneous anisotropic scatterers. It is shown that the hybrid method can achieve both higher reconstruction accuracy and lower computational cost compared with direct BIM inversion. Meanwhile, its antinoise ability is also tested.