An Attention-Based System for Damage Assessment Using Satellite Imagery

Hao, Hanxiang; Baireddy, Sriram; Bartusiak, Emily R.; Konz, Latisha; LaTourette, Kevin J.; Gribbons, Michael; Chan, Moses W.; Comer, Mary L.; Delp, Edward J.

doi:10.48550/arxiv.2004.06643

Cited by 7 publications

(12 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…One specific area that has garnered significant attention in computer vision and satellite imagery is building damage assessment. Recent works have studied semantic building segmentation [12,11] and cross-region transfer learning to avoid overfitting [25]. Furthermore, [9] presents a semi-supervised approach.…”

Section: Related Workmentioning

confidence: 99%

“…Because it is relatively difficult to obtain damage assessment and other details from on the ground in a timely manner, satellite imagery has gained popularity in being used to analyze these types of situations. Deep neural networks (DNNs) have been used to locate and classify building damage within satellite imagery [12,11,25,9]. However, the current literature is limited in the interpretability of what exactly these neural networks are learning and what is most useful in assessing building damage.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Interpretability in Convolutional Neural Networks for Building Damage Classification in Satellite Imagery

Chen¹

2021

Preprint

View full text Add to dashboard Cite

Natural disasters ravage the world's cities, valleys, and shores on a monthly basis. Having precise and efficient mechanisms for assessing infrastructure damage is essential to channel resources and minimize the loss of life. Using a dataset that includes labeled pre- and post- disaster satellite imagery, we train multiple convolutional neural networks to assess building damage on a per-building basis. In order to investigate how to best classify building damage, we present a highly interpretable deep-learning methodology that seeks to explicitly convey the most useful information required to train an accurate classification model. We also delve into which loss functions best optimize these models. Our findings include that ordinal-cross entropy loss is the most optimal loss function to use and that including the type of disaster that caused the damage in combination with a pre- and post-disaster image best predicts the level of damage caused. Our research seeks to computationally contribute to aiding in this ongoing and growing humanitarian crisis, heightened by climate change.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Interpretability in Convolutional Neural Networks for Building Damage Classification in Satellite Imagery

Chen¹

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The same is true for the SpaceNet Challenge Series for building footprint extraction, where the default splits created by the challenge's utilities contain overlapping locations [69]. Subsequent studies have shown that performance drops drastically when applying a trained building damage classifier to an unseen location, even within the same region [29,68]. As another example requiring spatial generalization, until very recently, all studies of animal species classification on camera trap images were split across sequences of images but not across locations.…”

Section: Creating Training and Test Sets With The Application Scenari...mentioning

confidence: 99%

Becoming Good at AI for Good

Kshirsagar

Robinson

Yang

et al. 2021

Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society

View full text Add to dashboard Cite

AI for good (AI4G) projects involve developing and applying artificial intelligence (AI) based solutions to further goals in areas such as sustainability, health, humanitarian aid, and social justice. Developing and deploying such solutions must be done in collaboration with partners who are experts in the domain in question and who already have experience in making progress towards such goals. Based on our experiences, we detail the different aspects of this type of collaboration broken down into four high-level categories: communication, data, modeling, and impact, and distill eleven takeaways to guide such projects in the future. We briefly describe two case studies to illustrate how some of these takeaways were applied in practice during our past collaborations. CCS CONCEPTS• General and reference → Surveys and overviews; • Humancentered computing → Collaborative and social computing; • Social and professional topics → Sustainability.

show abstract

“…Standard practice is to use a two-branch CNN-like architecture with feature fusion schemes. For instance, Hao et al [13] concatenated the features from pre-and post-disaster images and fed them into a CNN-based framework. Gupta et al [14] developed a framework that uses the difference between pre-and postdisaster features as input of the deep neural network, which is denoted as RescueNet.…”

Section: Introductionmentioning

confidence: 99%

“…More recently, attention mechanism has been used with deep neural networks for remote sensing images processing [15], [16], which is a strategy of allocating larger weights to informative parts of an image/feature. For instance, the average pooling is applied to extract the channel attention information for remote sensing image segmentation [17], and a non-local-based attention module is utilized in the arXiv:2105.07364v1 [cs.CV] 16 May 2021 network to explore the spatial relations of satellite images, in [13].…”

Section: Introductionmentioning

confidence: 99%

BDANet: Multiscale Convolutional Neural Network With Cross-Directional Attention for Building Damage Assessment From Satellite Images

Shen

Zhu

Yang

et al. 2022

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Fast and effective responses are required when a natural disaster (e.g., earthquake, hurricane, etc.) strikes. Building damage assessment from satellite imagery is critical before relief effort is deployed. With a pair of pre-and postdisaster satellite images, building damage assessment aims at predicting the extent of damage to buildings. With the powerful ability of feature representation, deep neural networks have been successfully applied to building damage assessment. Most existing works simply concatenate pre-and post-disaster images as input of a deep neural network without considering their correlations. In this paper, we propose a novel two-stage convolutional neural network for Building Damage Assessment, called BDANet. In the first stage, a U-Net is used to extract the locations of buildings. Then the network weights from the first stage are shared in the second stage for building damage assessment. In the second stage, a two-branch multi-scale U-Net is employed as backbone, where pre-and post-disaster images are fed into the network separately. A cross-directional attention module is proposed to explore the correlations between pre-and post-disaster images. Moreover, CutMix data augmentation is exploited to tackle the challenge of difficult classes. The proposed method achieves stateof-the-art performance on a large-scale dataset -xBD. The code is available at https://github.com/ShaneShen/BDANet-Building-Damage-Assessment.

show abstract

An Attention-Based System for Damage Assessment Using Satellite Imagery

Cited by 7 publications

References 24 publications

Interpretability in Convolutional Neural Networks for Building Damage Classification in Satellite Imagery

Interpretability in Convolutional Neural Networks for Building Damage Classification in Satellite Imagery

Becoming Good at AI for Good

BDANet: Multiscale Convolutional Neural Network With Cross-Directional Attention for Building Damage Assessment From Satellite Images

Contact Info

Product

Resources

About