After natural disasters, it’s important to understand which areas suffered the most damage in order to prioritize relief efforts. Often times damage assessment maps are created by volunteers with the Humanitarian Open Street Map Team who compare pre and post satellite imagery and hand label buildings with their evaluation of damage. However these maps are time and labor intensive to create, and not always accurate. My goal was to create a model that could more quickly and more accurately identify the hardest hit areas in order to better target disaster relief. Using satellite imagery before and after Typhoon Haiyan in the Philippines, I built a neural network to detect damaged buildings. Using the predictions from the model, I created density maps of damage, illustrating priority areas for relief efforts.
Libraries: Keras + TensorFlow, numpy, pandas, sklearn, rasterio, geopandas, shapely, opencv, matplotlib, seaborn
Methods: Deep learning, classification (supervised learning)