Calamity classification in satellite and drone images using CNN and Transfer Learning

Every year, natural disasters strike, wreaking havoc on lives, property, and the environment forever. They are unstoppable phenomena. In recent years, remote sensing imaging analysis has become increasingly important for the purpose of identifying and monitoring natural disasters in the context of climate change and environmental surveillance. The ability to convey a large amount of information in a single piece while still capturing the status of the underlying ground is one advantage of aerial or satellite imaging. Remotely sensed imagery has become increasingly important for environmental and climate monitoring in recent years, particularly for the detection and management of natural disasters. Using a convolutional neural network (CNN) to better extract the catastrophe characteristic, we propose autonomous natural disaster identification in this paper. CNN is resistant to shadows and is able to accurately determine the characteristics of a disaster. It is also able to avoid operator error, which can have an impact on the efficiency of disaster relief efforts. The images used as the model’s training data are divided into four categories: floods, cyclones, wildfires, and earthquakes. The input for disaster detection is either a live video stream or a video that has already been recorded. The study demonstrated that transfer learning can be used to accurately recognize natural disasters. We hope that the findings of this study will lead to the creation of monitoring or surveillance systems that are capable of accurately recognizing natural disasters in real-time. Based on the encouraging results, the suggested method may contribute to our understanding of deep learning’s role in catastrophe detection.