Disaster Damage Assessment Using Deep Learning and Satellite Imagery

This paper focuses on deep learning strategies for the assessment of satellite images for the overall assessment of disaster. The study primarily examines the ability to correctly identify those places that can be affected by various classes of natural disasters. By imbuing a wide range of satellite images with seamless integration, a new disaster detection system is designed assisted by a set of models, a prime example of which includes the Convolutional Neural Networks (CNNs). The above-mentioned detection system has demonstrated competency in the semantic segmentation and examination of satellite data of increased interest in both urban and countryside vistas. In the context of a city, the CNN model, supported by three advanced convolutional layers, max-pooling layers, and a double fully connected layer configuration, was painstakingly trained on an unparalleled dataset. This dataset consists of thousands of unique image patches before and after disastrous events. These represent various disastrous events that happened all over the world, thus enabling direct and comparative consideration of the pre- and post-disaster landscape. The procedures that are reported here can raise the level of performance and reliability of the practices in the Disaster Management field. It has presented an approach to analyze disaster (effect) efficiently and in depth considering recent technology on satellite image.