The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVI-3/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 133–140, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-133-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVI-3/W1-2022, 133–140, 2022
https://doi.org/10.5194/isprs-archives-XLVI-3-W1-2022-133-2022
 
22 Apr 2022
22 Apr 2022

ASSESSMENT OF BUILDINGS AND ELECTRICAL FACILITIES DAMAGED BY FLOOD AND EARTHQUAKE FROM SATELLITE IMAGERY

Y. Ma1, F. Zhou1, G. Wen1, H. Gen1, R. Huang1, G. Liu2, and L. Pei2 Y. Ma et al.
  • 1Electric Power Research Institute, Yunnan Power Grid Company ltd., Kunming, China
  • 2School of Electronic Information and Electrical Engineering, Shanghai Jiao Tong University, Shanghai, China

Keywords: Damage Assessment, Natural Disasters, Satellite Imagery, Buildings and Electrical Facilities, xBD Dataset, Building Localization

Abstract. Natural disasters cause considerable losses to people’s lives and property. Satellite images can provide crucial information of the affected areas for the first time, conducive to relieving the people in disaster and reducing the economic loss. However, the traditional satellite image analysis method based on manual processing drains workforce and material resources, which slowed the government’s response to the disaster. Aiming at the natural disasters like floods and earthquakes that often happen in the south of China, we propose a dual-stage damage assessment method based on LEDNet and ResNet. Our method detects the changes between the satellite images captured before and after a disaster of the same area, segments the buildings, and evaluates the damage level of affected buildings. In addition, we calculate influence maps based on the damage scale to the building and estimate the damage situation for electrical facilities. We used images related to earthquakes and floods in the xBD dataset to train the network model. Moreover, qualitative and quantitative evaluations demonstrated that our method has higher accuracy than the xBD baseline.