A Novel Deep Learning Framework Approach for Natural Calamities Detection

The term "natural calamities" has now emerged to be a global threat all over the world because of their steady mayhem and chaos in the form of human and economic loss. As these catastrophic events increase in severity and destruction, their classification based on their wide diversity is an important task from the disaster management aspect. However, it is yet an unpursued and unexamined topic in the branch of computer vision. This paper proposes a novel framework to detect natural calamities and thereby classify them in accordance with their class. A miscellany of ten different natural disasters, namely avalanche, cyclone, drought, earthquake, landslide, thunderstorm, tsunami, tornado, wildfire, and volcano was considered for classification. The framework solely relies on a hybrid of the convolutional neural network (CNN) for feature extraction. Because of the unavailability of a scrupulous dataset, a new dataset was constructed for testing the capability of the framework. The model performs exceptionally better when compared with other state-of-the-art algorithms (SVM, KNN, RF, and CNN + SVM). The results demonstrated that the model performs considerably better than average human intelligence in terms of different calamity and is able to recognize different classes of natural disasters with an accuracy of 82.23%.