text
stringlengths
0
820
[67] X. Wang, R. Girshick, A. Gupta, and K. He. Non-local neu-
ral networks. In 2018 IEEE/CVF Conference on Computer
Vision and Pattern Recognition , pages 7794–7803, 2018. 3
[68] Yuqing Wang, Zhaoliang Xu, Xinlong Wang, Chunhua Shen,
Baoshan Cheng, Hao Shen, and Huaxia Xia. End-to-end
video instance segmentation with transformers. In 2021
IEEE/CVF Conference on Computer Vision and Pattern
Recognition (CVPR) , pages 8737–8746, 2021. 3
[69] Li Yuan, Yunpeng Chen, Tao Wang, Weihao Yu, Yujun Shi,
Zi-Hang Jiang, Francis E.H. Tay, Jiashi Feng, and Shuicheng
Yan. Tokens-to-token vit: Training vision transformers from
scratch on imagenet. In Proceedings of the IEEE/CVF In-
ternational Conference on Computer Vision (ICCV) , pages
558–567, October 2021. 3
[70] Peng Yue, Boyi Shangguan, Lei Hu, Liangcun Jiang, Chenx-
iao Zhang, Zhipeng Cao, and Yinyin Pan. Towards a train-
ing data model for artificial intelligence in earth observation.
International Journal of Geographical Information Science ,
36(11):2113–2137, 2022. 1
[71] Zixiao Zhang, Xiaoqiang Lu, Guojin Cao, Yuting Yang,
Licheng Jiao, and Fang Liu. Vit-yolo:transformer-based
yolo for object detection. In 2021 IEEE/CVF International
Conference on Computer Vision Workshops (ICCVW) , pages
2799–2808, 2021. 3
[72] Bolei Zhou, Alex Andonian, Aude Oliva, and Antonio Tor-
ralba. Temporal relational reasoning in videos. In Pro-
ceedings of the European Conference on Computer Vision
(ECCV) , September 2018. 5
10428
On the Deployment of Post-Disaster Building
Damage Assessment Tools using Satellite Imagery:
A Deep Learning Approach
Shahrzad Gholami1, Caleb Robinson1, Anthony Ortiz1, Siyu Yang1, Jacopo Margutti2,
Cameron Birge1, Rahul Dodhia1, Juan Lavista Ferres1
1AI for Good Research Lab, Microsoft, Redmond, USA,
2510 an initiative of the Netherlands Red Cross, The Hague, The Netherlands
Abstract —Natural disasters frequency is growing globally.
Every year 350 million people are affected and billions of
dollars of damage is incurred. Providing timely and appropriate
humanitarian interventions like shelters, medical aid, and food to
affected communities are challenging problems. AI frameworks
can help support existing efforts in solving these problems in
various ways. In this study, we propose using high-resolution
satellite imagery from before and after disasters to develop a
convolutional neural network model for localizing buildings and
scoring their damage level. We categorize damage to buildings
into four levels, spanning from not damaged to destroyed, based
on the xView2 dataset’s scale. Due to the emergency nature
of disaster response efforts, the value of automating damage
assessment lies primarily in the inference speed, rather than
accuracy. We show that our proposed solution works three
times faster than the fastest xView2 challenge winning solution
and over 50 times faster than the slowest first place solution,
which indicates a significant improvement from an operational
viewpoint. Our proposed model achieves a pixel-wise F1 score
of 0.74 for the building localization and a pixel-wise harmonic
F1 score of 0.6 for damage classification and uses a simpler
architecture compared to other studies. Additionally, we develop
a web-based visualizer that can display the before and after
imagery along with the model’s building damage predictions on
a custom map. This study has been collaboratively conducted to
empower a humanitarian organization as the stakeholder, that
plans to deploy and assess the model along with the visualizer
for their disaster response efforts in the field.
Index Terms —satellite imagery datasets, neural networks,
image segmentation, building damage classification, natural
disasters, humanitarian action
I. I NTRODUCTION
Natural disasters affect 350 million people each year causing
billions of dollars in damage and were the main driver of
hunger for 29 million people in 2021 [1]. Providing timely
humanitarian aid to affected communities is increasingly
challenging due to the growing frequency and severity of
such events [2]. Impact assessment of natural disasters in
a short time frame is a crucial step in emergency response
efforts as it helps first responders allocate resources effectively.
For example, dispatching aid, sending shelters, and allocating
building material for reconstruction can be more efficient with
estimates of where damaged buildings are, and how badly
damaged they are.Microsoft AI for Good/Humanitarian Action has collab-
orated with Netherlands Red Cross to use high-resolution
satellite imagery from before and after natural disasters,
delineated in the publicly available xBD dataset, to develop
an end-to-end Siamese convolutional neural network that can
localize buildings and score their damage level. Such a model
is trained on historical disaster data and then applied on
demand to identify damaged buildings during future disasters.
Such AI and data-driven decision-aid tools can empower
humanitarian organizations to take more informed actions
at the time of disaster and allocate their resources more
strategically during their field deployments. Throughout the
course of our collaboration, extensive deployment experience
shared by field experts and their valuable perspective as
a stakeholder were instrumental in informing our empirical
analysis of the model pipeline and will be vital in future
assessments of the model performance in the fields when
actual disasters happen.
In 2019, the xView2 challenge and the xBD dataset were
announced at the Computer Vision for Global Challenges