Conventional microwave imaging methods, including synthetic aperture radar (SAR), suffer from slow data acquisition and high hardware complexity, limiting their suitability for real-time imaging, particularly for electrically large scenes. To address these challenges, computational microwave imaging (CMI) is considered an alternative technique. CMI-based systems successfully achieve a physical layer compression, but are still limited by the computational burden in the signal processing layer. To improve the efficiency of CMI-based systems, deep learning techniques have been explored to reduce the computational burden associated with conventional methods.In this thesis, a series of deep learning–based methodologies are developed to enhance the overall efficiency of CMI. These include two ways of transfer function prediction, a deep learning approach for multi-tasking in CMI application, and a deep learning-based reconstruction method capable of handling obstructed measurement conditions. Together, these approaches provide a coherent framework that directly addresses the aforementioned challenges and advances the state of the art in CMI reconstruction.The proposed approaches significantly reduce the computational effort associated with transfer-function estimation, remove the need for labor-intensive near-field measurements, enable accurate joint reconstruction and recognition, and effectively recover target information in obstructed scenarios.Overall, the contributions of this thesis establish a deep learning–driven foundation for efficient, scalable, and resilient CMI, supporting its advancement toward real-time deployment in practical environments.
- Computational Microwave Imaging
- deep learning
- generative adversarial network
- sensing matrix
Deep learning in compressive radar imaging and systems
Zhang, J. (Author). Jul 2026
Student thesis: Doctoral Thesis › Thesis with Publications