This thesis investigates utilizing deep learning techniques to enhance Microwave Imaging (MWI). MWI has been used for a broad range of applications, from medical imaging and security to industrial engineering, subsurface prospection and stored-grain monitoring. Like other applications, medical imaging and stored-grain monitoring, which are of interest in this thesis, have recently received a lot of attention. While traditional MWI approaches have made great progress in recent years, there are still many fundamental challenges. Lower resolution compared to other imaging modalities, as well as the many reconstruction artifacts are the main challenges for MWI. In the first part of this thesis, in order to remove reconstruction artifacts and enhance the tomographic reconstructions of the complex-valued permittivity obtained using traditional MWI techniques in the breast imaging application, a 2D Convolutional Neural Network (CNN) based on the U-Net architecture is proposed. The proposed CNN takes in images obtained using the Contrast-Source Inversion (CSI) technique and attempts to produce the true image of the permittivity. The results demonstrate that the trained CNN not only removes common artifacts in CSI reconstructions, but also improves tumor detection performance.

Considering these promising results on 2D images and with the hope of being able to deal with 3D real-world microwave imaging problems, we expand the U-Net architecture to enhance the 3D reconstructed breast images as the second part of this thesis. A 3D CNN was developed and trained to improve the tomographic reconstructions. Significantly, the results show that while the network is trained only with 3D images obtained synthetically, it works well with 3D images obtained from both synthetic and experimental data.

Finally, a novel CNN architecture is developed to replace the inversion technique itself. The proposed architecture is a multi-branch deep convolutional neural network designed to solve electromagnetic inverse scattering problems. Inspired by traditional iterative techniques for solving the electromagnetic inverse scattering problems, the proposed CNN architecture takes in scattered-field data and prior information to produce 3D images of grain moisture content in the stored-grain application. The proposed CNN's input is of two types:​ a complex-valued vector of scattered-field data, and a 3D image of the background moisture distribution as prior information. Thus, a multi-branch architecture consisting of decoder-only, and encoder-decoder, convolutional branches is proposed. The reconstructed moisture distribution images produced by the proposed CNN show that the network can not only reconstruct the 3D moisture distribution images directly from measured scattered-field data for high contrast objects-of-interest, but it also achieves a higher imaging quality compared with traditional inversion techniques in microwave imaging.