Locating Refugee Presence using Satellite Imagery

Lebanon

Client Brief

Because of the war, thousands of Syrians have fled their homeland to seek refuge in Lebanon
Many of them live in informal rural settlements. Locating them and estimating their number is critical to the international community and actors wishing to support them

Solution

Use advanced deep learning methods to identify the tents scattered across Lebanon on high-resolution satellite imagery
Check the findings in the field with local consultants
Based on this identification, follow the evolution (where ? how many ?) of the informal refugee settlements and assess their access to infrastructure (roads, markets)

Results

Training of a Convolutional Neural Network with roughly 3,000 images distributed between tents/not tents areas
Testing the algorithm on a batch of 1,000 other labeled images
Accuracy: 90% of the tents are correctly identified