Disaster Risk Mapping from Aerial Imagery Using Deep Learning Techniques

In regions prone to natural disasters, the buildings must follow specific construction standards to avoid demolition. One of the factors that predict the risk of damage is the roof material. This paper investigates the performance of various deep convolutional neural network architectures to classify buildings based on roof material from aerial drone imagery. We also propose a method that is an ensemble of ResNet, ResNeXt, and EfficientNet variants of convolutional neural networks, which performed the best in our experiments. We obtained a log loss value as low as 0.4373 using the proposed method. Therefore, the proposed method can be used to perform an accurate classification of roof material using aerial drone imagery.