Rapid Post-earthquake Damage Detection Using UAV Imagery and YOLO: A Case Study from the Kahramanmaras Earthquake
Journal of the Indian Society of Remote Sensing, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s12524-026-02548-1
- Dergi Adı: Journal of the Indian Society of Remote Sensing
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Geobase, INSPEC, Natural Science Collection (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Building damage assessment, Disaster management, Geospatial dataset, Post-earthquake reconnaissance, UAVimagery, YOLO
- Hacettepe Üniversitesi Adresli: Evet
Özet
Earthquakes cause severe damage to urban infrastructure, making rapid damage assessment crucial for emergency response and recovery planning. Following the February 6th Kahramanmaras earthquake, UAV imagery from atlas.gov.tr was used to develop a dataset for detecting collapsed buildings. YOLOv9 and YOLOv12, advanced object detection models, were trained to classify different building states, focusing on collapsed structures. To enhance model performance and generalization, cross-validation techniques were applied across data subsets, and image mirroring was used to improve image diversity. The model demonstrated strong detection performance, achieving a peak overall precision of 0.965 and an F1-score of 0.937, with particularly reliable localization quality reflected by an mAP50-95 of 0.805. For collapsed buildings, the model reached a precision of 0.954 and an F1-score of 0.899, demonstrating its effectiveness for post-earthquake damage detection using YOLOv9-C. Integrating UAV imagery with deep learning enables rapid assessment, supporting emergency response, urban planning, and disaster recovery. This approach enhances disaster management by ensuring accurate structural damage detection and aiding informed decision-making. The results highlight the potential of deep learning in automating large-scale damage assessment.