Development of Automated Damage Detection Method for Urban Areas Using High-resolution Satellite Images

利用QuickBird（QB）在2004年12月26日伊朗巴姆地震前后获得的巴姆市高分辨率卫星图像，进行了建筑物损坏的视觉解释。通过对比地震前后的泛锐化QB图像，根据欧洲宏观地震分级（EMS-98）识别出被碎片包围的建筑物（3级）、部分倒塌的建筑物（4级）和完全倒塌的建筑物（5级）。共有12063栋建筑被分类；五年级4951人，四年级1700人。与Hisada et al.（2004）详细的实地调查数据相比，视觉损伤解释似乎给出了相当准确的结果。然而，由于0.6m分辨率的近垂直图像的限制，观察到一定的遗漏误差。基于事件后灰度图像计算边缘特征，采用IKONOS和QB图像进行自动损伤检测。由于事件后的QB图像与IKONOS图像相比色调有些不自然，因此在自动损伤检测之前对事件后的QB图像进行了颜色修改。IKONOS和QB图像所得到的受损区域与目视检查结果和实地调查数据相当吻合。将自动损伤检测方法应用于事件前QB图像，检测误差有限。在此基础上，验证了高分辨率光学卫星图像在建筑物损伤检测中的准确性和实用性。需要对不同灾害的不同图像进行进一步的验证。

Using high-resolution satellite images of Bam City acquired by QuickBird(QB) before and after the 26 December, 2004 Bam, Iran earthquake, visual interpretation of building damage was carried out. Comparing the pre-event and post-event pan-sharpened QB images, buildings surrounded by debris (Grade 3), partially collapsed buildings (Grade 4), and totally collapsed buildings (Grade 5) were identified based on the European Macroseismic Scale (EMS-98). A total of 12,063 buildings were classified ; 4,951 as Grade 5 and 1,700 as Grade 4. Comparing with detailed field survey data by Hisada et al. (2004), the visual damage interpretation seems to give reasonably accurate results. However some amount of omission error was observed due to the limitation of near-vertical images with 0.6m resolution.Post-event IKONOS and QB images were employed in an automated damage detection method, which uses the edge characteristics calculated from post-event gray scale images. Since the post-event QB image shows a bit unnatural color tone comparing with that of the IKONOS image, the color modification was applied to the post-event QB image before the automated damage detection. The resultant damaged areas from the IKONOS and QB images agreed reasonably well with the visual inspection results and the field survey data. The automated damage detection method was also applied to the pre-event QB image, and the commission errors were found to be limited.Based on these studies, the accuracy and usefulness of the high-resolution optical satellite images in building damage detection was demonstrated. A further validation of the proposed method is required for various images from different disasters.