
上周六,日出后不久,一颗卫星将远程摄像机对准了缅甸的曼德勒市,该市距离周五那场7.7级地震的震中不远,那场地震给缅甸的第二大城市带来了严重破坏。
此次卫星的任务是拍摄图像,这些图像将与人工智能技术相结合,助力救援组织快速评估出建筑物倒塌或严重受损的数量,进而精准定位救援人员最迫切需要前往的区域。
起初,这种高科技的计算机视觉方法并未奏效。
微软首席数据科学家胡安·拉维斯塔·费雷斯(Juan Lavista Ferres)表示:“在这次特定事件中,我们面临的最大障碍来自云层。这项技术无法穿透云层。”
云层最终散去,总部位于旧金山的行星实验室(Planet Labs)调动另一颗卫星,耗费数小时完成了航拍照片的拍摄工作,并将其传输给微软公益性的AI for Good Lab。彼时,华盛顿州雷德蒙德的微软总部时间已接近周五晚上11点,一群微软员工已准备就绪,等待接收数据。
AI for Good Lab此前已开展过人工智能辅助灾害评估工作,比如追踪2023年利比亚的特大洪灾,以及今年洛杉矶的野火情况。不过,此次面对缅甸曼德勒市的情况,他们没有依赖那种能够处理任何视觉数据的通用人工智能计算机视觉模型,而是必须针对曼德勒市的独特状况构建定制化模型。
拉维斯塔·费雷斯说:“地球各个区域的地理环境差异太大,自然灾害也各不相同,我们从卫星获取的图像在不同情形下同样存在显著差异,这使得统一的模型难以普适。”他举例说明,火灾的蔓延方式相对可预测,但“地震会波及整个城市”,在震后初期很难迅速确定哪些区域迫切需要援助。
人工智能分析的结果显示,曼德勒市有515栋建筑物的受损程度达到80%至100%,另有1524座建筑物的受损程度在20%至80%之间。这一数据不仅凸显了这场灾难波及范围之广、破坏程度之重,同样重要的是,它还有助于确定具体的受损地点。
拉维斯塔·费雷斯说:“这对地面救援团队而言是至关重要的信息。”
微软提醒称,该报告“应作为初步指导,要全面了解情况还需进行实地核查”。但与此同时,这家科技公司已将分析结果分享给了红十字会等救援组织。
行星实验室表示,自上周五地震发生以来,其部署在地球轨道上的15颗卫星已对缅甸和泰国大约12个关键地点进行了拍摄。(财富中文网)
译者:中慧言-王芳
上周六,日出后不久,一颗卫星将远程摄像机对准了缅甸的曼德勒市,该市距离周五那场7.7级地震的震中不远,那场地震给缅甸的第二大城市带来了严重破坏。
此次卫星的任务是拍摄图像,这些图像将与人工智能技术相结合,助力救援组织快速评估出建筑物倒塌或严重受损的数量,进而精准定位救援人员最迫切需要前往的区域。
起初,这种高科技的计算机视觉方法并未奏效。
微软首席数据科学家胡安·拉维斯塔·费雷斯(Juan Lavista Ferres)表示:“在这次特定事件中,我们面临的最大障碍来自云层。这项技术无法穿透云层。”
云层最终散去,总部位于旧金山的行星实验室(Planet Labs)调动另一颗卫星,耗费数小时完成了航拍照片的拍摄工作,并将其传输给微软公益性的AI for Good Lab。彼时,华盛顿州雷德蒙德的微软总部时间已接近周五晚上11点,一群微软员工已准备就绪,等待接收数据。
AI for Good Lab此前已开展过人工智能辅助灾害评估工作,比如追踪2023年利比亚的特大洪灾,以及今年洛杉矶的野火情况。不过,此次面对缅甸曼德勒市的情况,他们没有依赖那种能够处理任何视觉数据的通用人工智能计算机视觉模型,而是必须针对曼德勒市的独特状况构建定制化模型。
拉维斯塔·费雷斯说:“地球各个区域的地理环境差异太大,自然灾害也各不相同,我们从卫星获取的图像在不同情形下同样存在显著差异,这使得统一的模型难以普适。”他举例说明,火灾的蔓延方式相对可预测,但“地震会波及整个城市”,在震后初期很难迅速确定哪些区域迫切需要援助。
人工智能分析的结果显示,曼德勒市有515栋建筑物的受损程度达到80%至100%,另有1524座建筑物的受损程度在20%至80%之间。这一数据不仅凸显了这场灾难波及范围之广、破坏程度之重,同样重要的是,它还有助于确定具体的受损地点。
拉维斯塔·费雷斯说:“这对地面救援团队而言是至关重要的信息。”
微软提醒称,该报告“应作为初步指导,要全面了解情况还需进行实地核查”。但与此同时,这家科技公司已将分析结果分享给了红十字会等救援组织。
行星实验室表示,自上周五地震发生以来,其部署在地球轨道上的15颗卫星已对缅甸和泰国大约12个关键地点进行了拍摄。(财富中文网)
译者:中慧言-王芳
Just after sunrise on Saturday, a satellite set its long-range camera on the city of Mandalay in Myanmar, not far from the epicenter of Friday’s 7.7 magnitude earthquake that devastated the Southeast Asian country’s second-largest city.
The mission was to capture images that, combined with artificial intelligence technology, could help relief organizations quickly assess how many buildings had collapsed or were heavily damaged and where helpers most needed to go.
At first, the high-tech computer vision approach wasn’t working.
“The biggest challenge in this particular case was the clouds,” said Microsoft’s chief data scientist, Juan Lavista Ferres. “There’s no way to see through clouds with this technology.”
The clouds eventually moved and it took a few more hours for another satellite from San Francisco-based Planet Labs to capture the aerial pictures and send them to Microsoft’s philanthropic AI for Good Lab. By then it was already about 11 p.m. Friday at Microsoft headquarters in Redmond, Washington. A group of Microsoft workers was ready and waiting for the data.
The AI for Good lab has done this kind of AI-assisted damage assessment before, tracking Libya’s catastrophic flooding in 2023 or this year’s wildfires in Los Angeles. But rather than rely on a standard AI computer vision model that could run any visual data, they had to build a customized version specific to Mandalay.
“The Earth is too different, the natural disasters are too different and the imagery we get from satellites is just too different to work in every situation,” Lavista Ferres said. For instance, he said, while fires spread in fairly predictable ways, “an earthquake touches the whole city” and it can be harder to know in the immediate aftermath where help is needed.
Once the AI analysis was complete, it showed 515 buildings in Mandalay with 80% to 100% damage and another 1,524 with between 20% and 80% damage. That showed the widespread gravity of the disaster, but, just as important, it helps pinpoint specific locations of damage.
“This is critical information for teams on the ground,” Lavista Ferres said.
Microsoft cautioned that it “should serve as a preliminary guide and will require on-the-ground verification for a complete understanding.” But in the meantime, the tech company has shared the analysis with aid groups such as the Red Cross.
Planet Labs says its satellites — it has 15 of them orbiting the Earth — have now photographed roughly a dozen locations in Myanmar and Thailand since Friday’s quake.