2021/02/16 by Fardin Ghorbani, Ghorbani, Fardin, Hossein Soleimani +1
Engineering · #Advanced SAR Imaging Techniques #Audio and Speech Processing (eess.AS) #FOS: Computer and information sciences #FOS: Electrical engineering #Geophysical Methods and Applications #Microwave Imaging and Scattering Analysis #Signal Processing (eess.SP) #Sound (cs.SD) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2102.07990
openalex publication_date 2021/02/16 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
This paper employed deep learning to do two-dimensional, multi-target locating in Through-the-Wall Radar under conditions where the wall is treated as a complex electromagnetic medium. We made five assumptions about the wall and two about the number of targets. There are two target modes available: single target and double targets. The wall scenarios include a homogeneous wall, a wall with an air gap, an inhomogeneous wall, an anisotropic wall, and an inhomogeneous-anisotropic wall. Target locating is accomplished through the use of a deep neural network technique. We constructed a dataset using the Python FDTD module and then modeled it using deep learning. Assuming the wall is a complex electromagnetic medium, we achieved 97.7% accuracy for single-target 2D locating and 94.1% accuracy for two-target locating. Additionally, we noticed a loss of 10% to 20% inaccuracy when noise was added at low SNRs, although this decrease dropped to less than 10% at high SNRs.