부산대학교 도서관

The extent of war-induced destruction in urban areas is critical information for international relief efforts, impact assessments and restoration decisions. However, precise geotargeting of zones with severe destruction is still a great challenge. Here we present a novel temporal-knowledge-guided detection scheme (TKDS) with a pixel-based transformer network (PtNet) for monitoring urban destruction using satellite imagery, applied to conflict zones in the Syrian civil war and the Russia–Ukraine conflict. Compared with state-of-the-art methods, the TKDS-PtNet model enhances war damage identification by 44.0 (72.5 versus 28.5) in the F1 score for six Syrian cities and 34.2 (83.5 versus 49.3) for four Ukrainian cities. The identified damaged buildings are further utilized to estimate the affected population and damage to critical infrastructures such as hospitals and schools in these areas. Our results demonstrate the high potential of a repeatable and relatively low-cost scheme for the near real-time monitoring of damage in urban areas resulting from wars, earthquakes or extreme weather events. Our findings underscore the crucial importance of taking action to stop the conflict and developing mechanisms to prevent present and future urban-related damage from military actions.
This study designs a new model based on medium-resolution satellite imagery to assess building damage from war, using the cases of Syria and Ukraine. It found that building damage has broader consequences for the population affected, especially when accounting for hospitals and schools.