快猫短视频

An algorithm that detects battle damage can compare satellite images to estimate聽the number of buildings destroyed and people affected by ongoing conflicts. It has been tested on satellite images of Ukraine and Gaza, and could be used to quickly guide relief and reconstruction efforts by humanitarian organisations.

The algorithm detected differences between intact and damaged buildings in 12 different cities by analysing radar signals from the European Space Agency鈥檚 Sentinel-1 satellites. These satellites typically capture a consistent return image when their radar bounces a signal off stationary objects such as buildings.

at University College London used a straightforward statistical test to compare the radar return signals for each building 鈥� and record battle damage based on any statistically significant changes.

This task can stump artificial intelligence because 鈥渢he models that are used for battle-damage detection are pretty bad at generalising to new geographies鈥�, says Ballinger.

Compared with images labelled by humans, the method was between 70 and 88 per cent accurate in detecting damaged buildings from a United Nations dataset of more than 633,000 buildings in 12 cities in Ukraine, Gaza, Syria and Iraq.

This tool could prove as accurate as costly AI models in development that often struggle to maintain accuracy in unfamiliar urban environments, says Ballinger. It is also cheaper because it relies on publicly available ESA satellite data.

鈥淰ery often we lack data and only the best-funded organisations can afford the satellite pictures we would require for doing real-time monitoring,鈥� says at the Institute for Economic Analysis in Spain.

Mueller thinks this approach is promising but says he wants to see how well the accuracy holds up 鈥渋n the wild鈥�. So far, Ballinger has created two dashboards through Google Earth for anyone to track the ongoing destruction in and 鈥� and he says that a UN team working on humanitarian efforts in Gaza has expressed interest in using the tool.