Hurricanes are incredibly dangerous and powerful storms, leaving many families dislocated and their homes destroyed.
What it Does
Hurricane Heroes is a program that uses a binary image classification to detect concentrations of damaged houses to help authorities decide what areas they should focus their efforts.
How We Built It
From xview2.org, we were able to find a dataset with 1024x1024 images to use as a training data of damaged and undamaged houses after a hurricane. Using satellite and flight imagery found on the National Geodetic Survey, we were able to map out a set of images and create a visual representation to show what areas of the hurricane were most damaged.
Our biggest challenges were definitely running flask and Azure. It was fairly new territory, which means the learning curve was high, but tinkering with the two programs was more complex than we had first anticipated.
The backbone of our product, the classifier, is built with Azure and although the classifier can be more accurate, as it stands right now, is pretty decent. In addition, we were able to link various programs, like Azure, python, and react so that they were able to talk to each other to create a robust code pipeline.
There are quite a few ways we can improve the accuracy of our classifier, both algorithmically and with more training data. In terms of functionality, we can expand to other natural disaster events such as tornadoes or earthquakes.