Collapsed buildings are usually among the places in which the highest number of human casualties are reported after a natural disaster. Therefore, detecting collapsed buildings immediately after an event and providing the information to first responders will expedite rescue operations and save human lives. The most efficient way to get this information immediately and in large scale is to use earth observing satellites. To this end, the U.S. Defense Innovation Unit Experimental (DIUx) and National Geospatial Intelligence Agency (NGA) have recently released a dataset known as "xView" that contains thousands of labeled VHR World View 3 optical satellite imagery scenes. One of the label classes is "demolished building" with more than a thousand instances. In this study, we are using the xView dataset to create a deep learning framework for detecting the collapsed buildings immediately after a natural hazard. We have used a U-net style fully convolutional neural network (CNN).
[1]
Aijun Chen,et al.
Regular Shape Similarity Index: A Novel Index for Accurate Extraction of Regular Objects From Remote Sensing Images
,
2015,
IEEE Transactions on Geoscience and Remote Sensing.
[2]
Thomas Brox,et al.
U-Net: Convolutional Networks for Biomedical Image Segmentation
,
2015,
MICCAI.
[3]
Yaroslav Bulatov,et al.
xView: Objects in Context in Overhead Imagery
,
2018,
ArXiv.
[4]
Hui Kong,et al.
A Generalized Laplacian of Gaussian Filter for Blob Detection and Its Applications
,
2013,
IEEE Transactions on Cybernetics.
[5]
Yang Shao,et al.
Detection of Urban Damage Using Remote Sensing and Machine Learning Algorithms: Revisiting the 2010 Haiti Earthquake
,
2016,
Remote. Sens..