The aftermath of the Haitian earthquake of 2010 involved a very detailed case study, which included one of the first uses of Artificial Intelligence to analyze a natural disaster. The main ideas which came into being involved the analysis of messages sent to the respective authorities, and the social media posts(eg:tweets) asking for help in the native Haitian creole. On similar lines, we plan to gather data from natural calamities, which include such SOSs, messages asking for help etc. and try to extract relevant information.
We use big data from sensor networks, social media (Tweets, SMSs), and from other available sources during a particular natural calamity, translate them, if necessary, and pre-process them, to fit our requirements.
We use RNNs, coupled with suitable classifiers to sort out the relevant information from the entire database, and acquire the exact locations where help was required, and the time when these messages were sent. With this, we give predictive insight over areas where help was required urgently, and find out the exact nature of help required.
We use deep learning on satellite imagery, with models such as Maximum Likelihood Classifiers(MLCs) and K-Means Classifiers, to predict the entire trajectory of the disaster and the areas with maximum damage. We use GEOBIA(Geographic Object Based Image Analysis) and GeoNode technologies to achieve this.
Institute- IIT Kharagpur
- Kunal Das
- Harshvardhan Srivastava
- Shobhit Mishra