Two-Stage Deep Learning Algorithm for Reconstructing the Permittivity Maps in Through the Wall Imaging Scenarios

Through-wall imaging (TWI) plays a crucial role in applications such as security, search and rescue operations, and infrastructure inspection. Traditional TWI methods face limitations such as low resolution, susceptibility to noise interference, and high computational demands. Ground-penetrating radar (GPR) has emerged as a promising alternative to conventional TWI technologies due to its portability, relatively rapid data processing, and potential for high-resolution imaging (M. Saad, A. Maali, M. S. Azzaz, A. Bouaraba, and M. Benssalah, “Development of an IR-UWB radar system for high-resolution through-wall imaging,” Progress In Electromagnetics Research C, vol. 124, pp. 81-96, 2022.). However, the raw data acquired using GPR is often significantly affected by noise, necessitating advanced signal processing techniques to extract meaningful information. Recent advancements in deep learning, particularly convolutional neural networks (CNNs), have shown remarkable effectiveness in processing noisy data and modeling complex nonlinear relationships (M. Küçükdemirci and A. Sarris, “GPR data processing and interpretation based on artificial intelligence approaches: future perspectives for archaeological prospection,” Remote Sensing, vol. 14, no. 14, July 2022). Integrating GPR with deep learning methods represents a powerful synergy to overcome the inherent limitations of the traditional GPR. To date, no study has demonstrated such synergy for denoising GPR data and accurately reconstructing the permittivity maps of objects behind walls in the context of TWI.