Deep Learning Meets InSAR for Infrastructure Monitoring: A Systematic Review of Models, Applications, and Challenges.

Area of Science:

• Geosciences
• Computer Science
• Civil Engineering

Background:

• Civil infrastructure monitoring is crucial due to aging assets and urban growth.
• Interferometric Synthetic Aperture Radar (InSAR) provides high-resolution surface deformation data.
• Deep Learning (DL) advancements are improving InSAR data analysis capabilities.

Purpose of the Study:

• To systematically review and synthesize the application of Deep Learning (DL) models in monitoring civil infrastructure using Interferometric Synthetic Aperture Radar (InSAR) data.
• To analyze DL architectures, their integration into the InSAR pipeline, and identify current trends and challenges.
• To outline future research opportunities for enhancing structural health monitoring.

Main Methods:

• Systematic literature review of 67 peer-reviewed articles (2020-Feb 2025).
• Analysis of Deep Learning (DL) architectures (LSTM, CNN, Transformers, hybrid models).
• Assessment of DL integration across InSAR monitoring stages (pre-processing, temporal analysis, segmentation, prediction, risk classification).

Main Results:

• LSTM and CNN models dominate current DL applications in InSAR for infrastructure monitoring.
• Limited exploration of DL for InSAR pre-processing tasks.
• Focus on urban and linear infrastructure monitoring, with emerging trends in hybrid architectures and data fusion.

Conclusions:

• Methodological challenges include data sparsity, low coherence, and lack of benchmarks.
• Emerging trends involve attention mechanisms, end-to-end pipelines, and integrating DL-InSAR into operational systems.
• Future research should focus on model explainability, diverse infrastructure types, and practical implementation.