Physics-Informed Anomaly Detection of Terrain Material Change in Radar Imagery

TL;DR

This work tackles terrain material-change detection in radar imagery by coupling a physics-informed forward model with a coherence-aware feature stack. A lightweight forward model maps per-pixel material parameters, such as permittivity $\varepsilon_r$, roughness $\sigma$, and correlation length $l_c$, to bi-temporal SLCs, while cross-epoch speckle decorrelation is governed by a coherence field $\gamma$ and expressed through $S^{(t)}_{p}=\sqrt{\sigma^{0,(t)}_{p}}\,\eta^{(t)}_{p}\,e^{j\phi^{(t)}_{p}}$ and $W_2=\gamma W_1+\sqrt{1-|\gamma|^2}W_\perp$. A physics-aware feature stack (including $I_1$, $I_2$, $R_{\log}$, texture, incidence, and $|\hat{\gamma}|$) feeds unsupervised detectors: RX with Tyler's robust estimator, Local-RX, Coherent Change Detection, and a compact autoencoder. Through 200 Monte Carlo trials across varied clutter (Gamma/$K$) and material changes, coherence-centric detectors (CCD and fused scores) significantly outperform intensity-based methods, with fusion approaching CCD performance and robustness from the Tyler estimator. The results illuminate when material changes manifest primarily through decorrelation and provide a reproducible framework linking EM properties to observable SAR cues, with practical implications for terrain monitoring in radar imagery. The study also notes the current limitations of single-frequency/polarization modeling and suggests extensions to PolSAR for broader applicability.

Abstract

In this paper we consider physics-informed detection of terrain material change in radar imagery (e.g., shifts in permittivity, roughness or moisture). We propose a lightweight electromagnetic (EM) forward model to simulate bi-temporal single-look complex (SLC) images from labelled material maps. On these data, we derive physics-aware feature stacks that include interferometric coherence, and evaluate unsupervised detectors: Reed-Xiaoli (RX)/Local-RX with robust scatter (Tyler's M-estimator), Coherent Change Detection (CCD), and a compact convolutional auto-encoder. Monte Carlo experiments sweep dielectric/roughness/moisture changes, number of looks and clutter regimes (gamma vs K-family) at fixed probability of false alarm. Results on synthetic but physically grounded scenes show that coherence and robust covariance markedly improve anomaly detection of material changes; a simple score-level fusion achieves the best F1 in heavy-tailed clutter.

Figures (3)

• Figure 1: Representative trial - feature maps. Left$\rightarrow$right: log-$I_1$, log-$I_2$, coherence magnitude $|\hat{\gamma}|$, ground truth (GT).
• Figure 2: Representative trial - detector score maps. Left$\rightarrow$right: RXrob, Local-RX, CCD ($1-|\hat{\gamma}|$).
• Figure 3: Representative trial. ROC computed on the same trials as Figs. \ref{['fig:qual']}\ref{['fig:detectors']}