Red Cross Mapping

Intel AI Lab, MILA, and CrowdAI conducted a joint data science collaboration to develop a deep learning image segmentation model to detect bridges in remote areas using satellite imagery. Working with the American Red Cross and its Missing Maps project, Uganda was identified as a starting point, given the country’s historical suffering from both community epidemics and seasonal weather events. A custom training dataset of high-resolution satellite imagery of Northern Uganda was used to train multiple models, tackling many challenges including variances in image angles, light and shadows, cloud cover, and seasonal shifts. The top-performing model was then run on satellite imagery from Southern Uganda, identifying 70 new bridges previously unseen to the model, cutting mapping time significantly.