关键词: 地震, 热红外数据, 无人机, 倒塌房屋

Abstract: Remote sensing is widely used for large-scale land surface information acquisition, such as land surface temperature estimation, land use and land cover classifications. Compared with in-situ field investigation, remote sensing have many advantages of efficiency and cost saving of human and material resources. As for the seismic risk reduction, the building area extraction, building height and other parameters acquisition, building structure type classification, and identification of post-earthquake building collapse, earthquake induced landslide, quake lake and other secondary disasters are the most important parts of remote sensing applications. The high- or mid-spatial resolution visible images of submeter to meters acquired from airborne and spaceborne instruments are widely used for damage building identifications, such as the April 14th Yushu earthquake in 2010. But the visible images cannot be used for nighttime land surface information acquisition. Thus, the active remote sensing techniques such as SAR or LiDAR are effective substitution for nighttime earthquake disaster information acquisitions. But the SAR and LiDAR data are not shown in traditional way which are similar with the visible images, and can only be interpreted by professional researchers. Note that some of the earthquakes are occurred during the nighttime, for example the July 28th Tangshan earthquake is occurred at the local time of 3:42. The nighttime earthquake disaster information acquisition is as important as the daytime. Considering the thermal infrared (TIR) sensor can receive thermal radiances emitted by surface features with the wavelength of 3 to 100 μm and independent on solar radiations, this study tries to used TIR data acquired during nighttime to identify earthquake building damage types. This study took the Beichuan earthquake site of the Wenchuan earthquake in 2008 as the research area. To acquire TIR image of the study area, the UAV of DJI M200 version 2 with the payload of ZENMUSE XT2 is used. The XT2 have two cameras, with one traditional RGB visible camera and one TIR camera with the center wavelength of 10 μm and spectral range of 7.5-13.5μm. The focal plane of XT2 have pixels of 4000×3000 for the visible camera and 640×512 for TIR camera. The pixel size of the TIR camera focal plane is 17 μm and the focal length is 13 mm. The M200 used in this study are flying at a height of 120 m above ground. Thus the spatial resolution of the corresponding TIR image is about 15.7 cm for vertical view condition. The TIR images of old town and new town of Beichuan county are acquired with M200. To evaluate the accuracy of damaged building identification using UAV TIR data, the visible images of the study areas are also acquired during the daytime and taken as reference. After mosaic the daytime visible images and nighttime TIR images, respectively, all the buildings in the study area are classified into three damage types by artificial interpretation: slight damage, moderate damage, and destruction. By comparing the damaged building types by visible remote sensing data as the true value, the confuse matrix are constructed for the TIR data estimated building damage types. It found that the UAV TIR remote sensing data can be used for the identification of damaged buildings at night, and the overall accuracy is 0.86, among which the user accuracy of the three damage types are all larger than 0.8. The mapping accuracy are 0.95, 0.67, and 0.84, respectively, for slight damage, moderate damage, and destruction types. The accuracy of moderate damage is 0.67. This is because the TIR image have only one channels and cannot show colors. Thus the moderate damage buildings are confused with trees and other features.

Key words: Earthquakes, thermal infrared image, UAV, collapsed buildings