Analysis of Social Media Imagery for Crisis Management Applications

Social media data holds immense potential for real-time disaster response. This study explores leveraging deep learning to automatically detect disaster-related information across various social media platforms. By analyzing the performance of different models in identifying relevant content, we aim to reduce information gathering delays and support timely rescue efforts. Faster information gathering translates to quick deployment of rescue teams, potentially saving lives and minimizing property damage. We evaluate these models on a benchmark dataset and explore the potential of combining them for even greater accuracy. Among the models, VGG16 achieved an accuracy of 81% in identifying disaster-related content. Additionally, exploring different fusion techniques for combining these models further improved accuracy to 83% with Hybrid Fusion. This research offers valuable insights for future exploration of deep learning techniques in disaster management.