SnapQuake: Damage Detection in Snapchat Videos for Earthquake Assessment

Social media data offers prompt access to situation-sensitive information, providing novel solutions for real-world crises. This study focuses on enhancing humanitarian event detection and assessment during catastrophes, notably the 2023 Syria earthquake, employing an innovative crisis event detection method with Snapchat. A refined listening model captured relevant Snapchats during targeted events, forming a dataset of 300 videos. Training the model for damage detection involved 11 videos each for damaged and undamaged scenarios; 4881 frames were extracted from these videos. MobileNetV2 was used for feature extraction, and a Bi-LSTM model was used for classification. The Snapquake model achieved an impressive 94.26% accuracy. Performance benchmarks highlighted the superiority of the SnapQuake architecture over MobileNetV2 and a hybrid ResNet34-Bi-LSTM model. This research underscores the potential of Snapquake as a valuable resource for timely disaster assessment in social media-driven disaster response, emphasizing its usability for assessing various crisis events with exact location information, thus broadening its application beyond earthquake assessment.