Cloud-based interactive database management suite integrated with deep learning-based annotation tool for landslide mapping

Abstract:

Understanding accurate spatial distribution of landslides is essential for landslide analysis, prediction and hazard mitigation. We present an interactive, user-friendly cloud-based pipeline to manage global landslide databases to channel user attention to suspected events. Our web-based platform maintains a database of multi-temporal satellite images, class of ecoregions, pixel-wise spatial distribution of landslide regions in satellite images and other metadata for geo-referenced landslide events. Geo-referenced Pre and Post event satellite images were collected and made available to users, along with metadata. Annotated labels along with the corresponding Pre and Post event satellite images were used to train multiple deep learning semantic segmentation models. Our pipeline is coupled with these trained models for automatic detection and localization of unmapped landslide areas to aid large-scale annotation. Two poorly cataloged landslide-affected regions were selected to test the capability of our cloud-based suite. The detected landslides were validated by expert labelers. The results indicated that our annotation tool was able to produce landslide maps with high precision, high rate of annotation and reduced human efforts. Based on the results, detailed landslide maps for the two selected regions were generated and statistical analysis was performed to evaluate the distribution of landslides. The results show that our high-precision landslide inventory shows its potential in facilitating landslide cataloging.