Classification of Natural Disaster Using Satellite & Drone Images with CNN Using Transfer Learning

Detection of natural disasters is a very crucial problem to mitigate the damage caused by them. Remote sensing technology can used in detection of certain disasters on an area. We have built a model to detect natural disasters like earthquake, cyclones, wildfires and floods using real time data from satellite and drones. The training data for the model consists of images classified into four classes namely earthquake, cyclones, wildfires and floods. The realtime video stream or a pre-recorded video is passed as the input for detection of calamities. Model takes each casing successively and measures it. For each frame, the probability of all the classes is calculated. Then class label with maximum probability is chosen. The class label is displayed on the top of the frame with a probability calculated by the model. If no disaster is predicted, it shows “normal” otherwise it will display the disaster predicted by the model.