As many of Houstonians, I was impacted by hurricane Harvey as well as my beloved once's. when I started to hear more and more accounts of rescuers being delayed getting to those who needed help, I started to search and think for a safer and faster rescue efforts.
I published a paper regarding this approach Detecting Damaged Buildings on Post-Hurricane Satellite Imagery based on Transfer Learning
What it does
Machine learning algorithms to identify and classify structural damages caused by floods through satellite images, these models can be used for various fields such as Structural Health Monitoring (SHM), vehicles, rails, and roads.
How we built it
Convolution Neural Network (CNN) is used to detect damage and no damage images form satellite imagery using different classifiers. The dataset used belongs to the hurricane Harvey that caused several damages in the Houston area. In addition, 1D-convolutional neural network (CNN) was used to look at the weekend joints of the frames in buildings. The model consisted of CNN with residual learning modules and multi-scale modules was applied to explore the mode shapes by extracting the damaged properties. A spatial relation was founded using Gate Recurrent Unit (GRU) and CNN to detect the damaged parts of the structural building.
Challenges we ran into
One of the main challenges is locating the dataset.
Accomplishments that we're proud of
The approach of using Convolution Neural Network (CNN) was successful
What we learned
Structural damages can be identified through Satellite images
What's next for Machine Learning For Visual Detection Of Floods And Damages
- Test the model using other data sources like Transtar camera pictures or drones
- Build an add-on model to process videos through multiple video sources identified by City Of Houston
- Enhance the solution to process real-time data and extrapolate key data points to use for forecasting and prediction.
Ultimately this solution would be part of emergency preparedness and a prevention tool.
Log in or sign up for Devpost to join the conversation.