Estimating the number and size of forcibly displaced populations is key for allocating resources and assistance during humanitarian emergencies.
Satellite image analysis is one of the most effective means to provide accurate data to understand the situation on the ground. The manual analysis of satellite imagery has thus far been the most reliable method of mapping structures in settlements built to house displaced populations. Although automated methods exist, they have proven insufficient for the complex high-resolution satellite imagery necessary for mapping structures in different conditions and locations.
When dealing with conflict and humanitarian scenarios, precision in satellite image analysis is key to supporting critical operations on the ground.
UNOSAT is a technology-intensive programme delivering imagery analysis and satellite solutions to relief and development organizations. UNOSAT and Global Pulse partnered to apply artificial intelligence frameworks that can help analysts to identify and count structures within settlements in high-resolution satellite imagery in a fraction of the time and cost that manual analysis requires.
The project developed multiple deep neural network architectures that were tested in several humanitarian crises in Uganda, South Sudan, Nigeria and Iraq. Global Pulse and UNOSAT will work in 2018 to refine the algorithm in order to increase precision and to include it into current operational pipelines.
Paper: Humanitarian applications of machine learning with remote sensing data: review and case study in refugee settlement mapping.
