Hurricane Disaster Detection

Oluwadamilola Akinola and Deondre Martin

Mentor: Dr. Rahnemoonfar

Department of Information Systems, UMBC

We all know that the damage left by hurricanes can be really devastating, people, property, and the progress of certain areas are put at high risk when bracing the storm. Even when people do evacuate they come back to see the destruction of their homes and communities. Though this is a natural part of life in which we cannot stop the forces of nature, the power of technology gives us an advantage. Artificial Intelligence is the study of computers mirroring human intelligence in machines. Deep learning is more specific, being how machines learn which is part of machine intelligence and Supervised learning being a step above this as well. Together, these concepts create a convolutional neural network, which is a densely connected network of layers that acts as a classifier. What is needed in a CNN are convolution layers that are used for image classification along with spatial invariance. Our project uses a convolutional neural network accompanying semantic segmentation, which is the process of linking each pixel in an image to a class label to identify different objects and things in an image. This technique is being used to predict areas that are most prone to damage from hurricanes. Our project holds great significance because it can be used for not just natural disasters but also regular storms that can affect communities as well. The data that can be gathered once this practice is more widespread can be used to create programs that can revolutionize our preparation and response to national disasters. With all this in mind we believe that this will make computer vision become more in our ever advancing future.