Skip to main content

Testing Machine Learning Algorithms for The Automatic Recognition of Conflict Damage in Syria


In the event of a disaster in an urban region, the rapid identification of areas which have been damaged and/or destroyed is extremely useful for human rights and crisis response actors.  In remotely-sensed imagery, visual identification of such change has long been considered the most reliable method of identification.  In this process, two such images, one from before and one after the incident, are compared to visually identify and confirm areas that have sustained damage. To decrease the time necessary for analysis of future images, this project tested the application of machine learning algorithms to automatically identify damaged features, using the features manually identified in previous AAAS damage assessments to train the algorithms.