Google use of Machine Learning for disaster management
Natural disasters like earthquakes, hurricanes, floods, and avalanches are not under our control, but surely we can take preventive measures so that the damages caused by these disasters become as less as possible. In this preventive measurement, Artificial Intelligence and Machine Learning are our true companions.
One of the major harms these disasters cause is damaging the buildings. Manual analysis of such damages would take many weeks, even for experts. But the data of such damages are required within 2–3 days when the most urgent decisions are made.
To reduce the impact of such disasters, Google has formed one Convolutional Neural Network-based deep learning model that takes the satellite images as inputs, processes them, and generates the building damage assessments. Google has developed this technology with the United Nations World Food Program (WFP) Innovation Accelerator partnership.
This complete process of damage assessment is divided into 2 steps:
- Detect the presence of buildings in the satellite images.
- Classification of building as damaged or not.
This CNN model takes two images (161x161x3) as inputs for a 50 m x 50 m ground footprint centered on a given building. One image corresponds to before disaster, and the other image corresponds to after the disaster. The model tries to analyze the difference between the two given images and produce the output of either 0 and 1. 0 corresponds to no damage, and 1 corresponds to damage. This model has achieved an accuracy of 70% on average.
Once the damaged building is identified, it can be repaired and made sustainable for later disasters.
Enjoy Learning! Enjoy Thinking! Enjoy Algorithms!