Multi-Satellite PS-InSAR and ML-Based Anomaly Detection for Bridge Monitoring

Abstract

Structural health monitoring (SHM) of aging bridges requires reliable methods to capture deformation behavior at multiple scales. Persistent Scatterer Interferometric Synthetic Aperture Radar (PS-InSAR) provides millimeter-level displacement measurements, but interpretation of these datasets remains challenging for slender structures. This study presents an anomaly detection framework for PS-InSAR displacement data combining temporal feature analysis and spatial structuring. Features such as velocity and thermal sensitivity are analyzed using an unsupervised Isolation Forest model, with SHapley Additive exPlanations (SHAP) used for interpretability. Persistent scatterers are projected onto the bridge axis and aggregated into span-scale zones, and a normalized relative anomaly density metric is introduced to enable cross-satellite comparison. The framework is applied to descending-pass Sentinel-1 and RADARSAT Constellation Mission datasets over the Victoria Bridge in Montreal. Results show consistent identification of key anomalous segments across datasets despite differences in spatial resolution. The proposed approach provides a structured and interpretable framework for PS-InSAR-based bridge monitoring.