Graph-CNNs for RF Imaging: Learning the Electric Field Integral Equations

TL;DR

This work targets RF imaging as an inverse-scattering problem and introduces a fast EFIE-based forward model to generate data for training. It presents a graph-attention driven DNN, GAT-Res-UNet, that encodes system geometry through a graph and uses residual convolutions and a UNet head to reconstruct binary DoI surfaces from received fields. Across two synthetic datasets, the proposed architecture achieves lower error and crisper reconstructions than baselines, and exhibits greater resilience to noise and reduced receiver counts. The results advocate for physics-informed, geometry-aware deep learning approaches to enable low-latency, high-resolution RF imaging in challenging operating conditions.

Abstract

Radio-Frequency (RF) imaging concerns the digital recreation of the surfaces of scene objects based on the scattered field at distributed receivers. To solve this difficult inverse scattering problems, data-driven methods are often employed that extract patterns from similar training examples, while offering minimal latency. In this paper, we first provide an approximate yet fast electromagnetic model, which is based on the electric field integral equations, for data generation, and subsequently propose a Deep Neural Network (DNN) architecture to learn the corresponding inverse model. A graph-attention backbone allows for the system geometry to be passed to the DNN, where residual convolutional layers extract features about the objects, while a UNet head performs the final image reconstruction. Our quantitative and qualitative evaluations on two synthetic data sets of different characteristics showcase the performance gains of thee proposed advanced architecture and its relative resilience to signal noise levels and various reception configurations.

Figures (6)

• Figure 1: GAT-Res-UNet architecture for RF imaging via inverse scattering.
• Figure 2: Visualization of the magnitude of the electric field $|E_{\rm r}(\boldsymbol{s})|$ in arbitrary units over the DoI in the presence of three perfect electrical conductors of side and diameter $1.2\lambda$, demonstrating the complex interactions captured by the employed EFIEs. Non-overlapping pulse functions with $\lambda/20$ spacing in between them were used as bases.
• Figure 3: Comparison of validation scores for the proposed GAT-Res-UNet and baseline methods on MNIST and SHAPES data sets (lower is better).
• Figure 4: Visual comparison of the reconstruction of three MNIST images.
• Figure 5: Visual comparison of the reconstruction of three SHAPES images.