InSAR-Guard: Lightweight AI Framework for Landslide Hazard Detection

Abstract

Landslides are a type of natural disaster that causes massive loss of life and property. Continuous observation of ground deformation is necessary for early warning and mitigation of risks. Interferometric Synthetic Aperture Radar (InSAR) technology facilitates large-scale, all-weather observation of ground deformation, but manual interpretation of InSAR images is a time-consuming process. To address these issues, this project proposes InSAR-Guard, a lightweight artificial intelligence framework for automated large-scale landslide detection using deep learning algorithms. A customized YOLOv8 model is developed to detect deformation patterns from InSAR images efficiently through a systematic preprocessing, annotation, and evaluation process. To improve the model’s sensitivity to small-scale deformation patterns that may occur before landslides, a Convolutional Neural Network (CNN) model is also developed and trained on deformation maps. A performance comparison between YOLOv8 and CNN models is conducted using accuracy, Intersection over Union (IoU), and detection validity as evaluation criteria to identify the best-performing model for monitoring purposes. The best-performing model is then incorporated into a real-time web-based dashboard with alert functionality for continuous monitoring and timely updates. The proposed framework offers a scalable, efficient, and accurate solution for automated landslide hazard detection and disaster readiness.