AUTHOR=Fu Lv , Zhang Qi , Wang Teng , Li Weile , Xu Qiang , Ge Daqing 

TITLE=Detecting slow-moving landslides using InSAR phase-gradient stacking and deep-learning network

JOURNAL=Frontiers in Environmental Science

VOLUME=Volume 10 - 2022

YEAR=2022

URL=https://www.frontiersin.org/journals/environmental-science/articles/10.3389/fenvs.2022.963322

DOI=10.3389/fenvs.2022.963322

ISSN=2296-665X

ABSTRACT=Landslide is one of the major geohazards that endangers human lives and properties. Recently, efforts have been made to use Synthetic Aperture Radar Interferometry (InSAR) for landslide monitoring. However, it is still difficult to effectively and automatically identify slow-moving landslides distributed in a large area due to phase unwrapping errors, troposphere turbulence and computational requirements. In this study, we develop a new approach combining phase-gradient stacking and deep-learning network based on YOLOv3, to automatically detect slow-moving landslides from large-scale interferograms. Using the Sentinel-1 SAR images acquired from 2014 to 2020, we develop a burst-based, phase-gradient stacking algorithm to sum up phase gradients in short-temporal-baseline interferograms along the azimuth and range directions. The stacked phase gradients clearly reveal the characteristics of localized surface deformation mainly caused by slow-moving landslides, avoiding the errors due to phase unwrapping and atmospheric effects. We then train the improved Attention-YOLOv3 network with the stacked phase-gradient maps of manually labelled landslides to achieve quick and automatic detection. We apply our method in an ~180,000km2 area of southwestern China and identify 3366 slow-moving landslides. By comparing the results with optical imagery and previously published landslides in this region, the proposed method can achieve the automatic detection in a large area precisely and efficiently. From the derived landslide density map, we find that most of the landslides are distributed along the three large rivers and their branches. In addition to some counties with known high-dense landslides, about 10 more with high landslide density are exposed, which should attract more attention to their geohazards risks. The presented application demonstrates the potential value of our newly developed method for slow-moving landslide detection in a large area, which can be employed before applying the more time-consuming time-series InSAR analysis.