Timezone: »

 
SAR-based landslide classification pretraining leads to better segmentation
Ragini Bal Mahesh · Ioannis Prapas · Wei Ji Leong · Vanessa Boehm · Edoardo Nemni · Freddie Kalaitzis · Siddha Ganju · Raul Ramos-Pollán

@
in The Fourth Workshop on AI for Humanitarian Assistance and Disaster Response »

Rapid assessment after a natural disaster is key for prioritizing emergency resources. In the case of landslides, rapid assessment involves determining the extent of the area affected and measuring the size and location of individual landslides. Synthetic Aperture Radar (SAR) is an active remote sensing technique that is unaffected by weather conditions. Deep Learning algorithms can be applied to SAR data, but, training them requires large labeled datasets. In the case of landslides, these datasets are laborious to produce for segmentation, and often they are not available for the specific region in which the event occurred. Here, we study how deep learning algorithms for landslide segmentation on SAR products can benefit from pretraining on a simpler task and from data from different regions. The method we explore consists of two training stages. First, we learn the task of identifying whether a SAR image contains any landslides or not. Then, we learn to segment in a sparsely labeled scenario where half of the data do not contain landslides. We test whether the inclusion of feature embeddings derived from stage-1 helps with landslide detection in stage-2. We find that it leads to minor improvements in the Area Under the Precision-Recall Curve, but also to a significantly lower false positive rate in areas without landslides and an improved estimate of the average number of landslide pixels in a chip. A more accurate pixel count allows to identify the most affected areas with higher confidence. This could be valuable in rapid response scenarios where prioritization of resources at a global scale is more important.

Author Information

Ragini Bal Mahesh (German Aerospace Center DLR)
Ioannis Prapas (University of Valencia, National Observatory of Athens)
Wei Ji Leong (The Ohio State University)
Vanessa Boehm (UC Berkeley)
Edoardo Nemni (United Nations Satellite Centre (UNOSAT))

Edoardo Nemni is a Machine Learning Researcher at the United Nations Institute of Training and Research Operational Satellite Application Programme (UNITAR-UNOSAT). His research focus lies on apply deep learning algorithms to satellite imagery for disaster response such as satellite-derived flood analysis, shelter mapping, building footprints, damage assessment, and more. His current project is FloodAI: an end-to-end fully automated pipeline whereby satellite images of flood-prone areas are automatically downloaded and processed to output disaster maps.

Freddie Kalaitzis (University of Oxford)
Freddie Kalaitzis

Freddie is a Senior Research Fellow at the Dept. of Computer Science, University of Oxford, investigating topics mainly in AI for Earth Observation. He is the principal investigator of OpenSR, a €1M government contract with ESA, to increase the safety of Super-Resolution technology for the Sentinel-2 archive. He is also an independent consultant, involved in projects where he leads teams in the Frontier Development Lab (FDL), a private-public partnership between NASA, SETI, and Trillium Technologies. His recent FDL projects were funded by NASA SMD to investigate the use of SAR imagery for disaster detection, and by the USGS to develop near-real-time water stream mapping from daily PlanetScope imagery. His most recent work is a survey on the State of AI for Earth Observation, in collaboration with Satellite Applications Catapult.

Siddha Ganju (Nvidia)
Raul Ramos-Pollán (Universidad de Antioquia)

More from the Same Authors