Disaster Scene Classification with Deep Learning: A Keras-Based Approach Utilizing Robotic Systems
Gautam Arora, R. M. Bhavadharini · 2024
This study presents an innovative technique that leverages convolutional neural networks (CNNs), an advanced computer vision methodology, to enhance the timely detection and classification of natural disasters. The objective is to develop a dependable model capable of categorizing various disaster- related photos and videos, such as cyclones, earthquakes, floods, and wildfires, by employing extensive datasets and sophisticated CNN architectures like Visual Geometry Group (VGG) including VGG16 and VGG19. A significant advancement introduced in this research is the seamless integration of robotics, enabling autonomous data collection and real-time decision-making at disaster sites. Through a technique encompassing model development and data preprocessing, the effectiveness of our approach in disaster scene classification is demonstrated. The findings exhibit remarkable performance: post 48 training epochs, the VGG19 model achieved an impressive 97% accuracy, while the VGG16 model achieved a commendable 93% accuracy. These outcomes underscore the potential of CNNs to revolutionize disaster management strategies and pave the way for swifter and more efficient responses to mitigate the impact of natural disasters on society.