Disaster Damage Assessment Using High Resolution Satellite Data and Deep Learning
||Disaster Damage Assessment Using High Resolution Satellite Data and Deep Learning|
||SNIC Small Compute|
||Dávid Kerekes <firstname.lastname@example.org>|
||Kungliga Tekniska högskolan|
||2020-11-12 – 2021-06-01|
In recent years, the world has the world has witnessed many natural disasters from flooding to wildfires. Earth observation satellite data can enable responders to identify the location, cause and severity of disaster damages. However, these natural disasters often impact large areas, and the manual process of searching through extensive imagery to pinpoint the location of and assess the amount of damage is slow and labor-intensive. Trained analysts who examine the images have to integrate their knowledge about an area’s geography, as well as the specific disaster’s conditions to score building damage.
In the project we are evaluating deep learning algorithms that can effectively process pre- and post-natural disaster satellite imagery to assess the severity of building damage. We are mainly using xBD, one of the largest and highest-quality publicly available collections of annotated high-resolution satellite imagery of damaged buildings, which consists of 850,000 building annotations across more than 45,000 km2 of imagery spanning 10+ countries and six different disaster types.
State of the art detection models for optical data make heavy use of deep convolutional neural networks, thus require large amounts of computational capacity both for their training and evaluation. We are hoping to leverage the GPU equipped nodes of the Tetralith cluster to find repeatable, high confidence methods for producing these models, thus making them applicable in a broad variety of future research and applications.