The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
L'identification rapide des dommages aux bâtiments joue un rôle essentiel dans les opérations de secours en cas de catastrophe, en particulier lorsque les ressources de secours sont limitées. Ces dernières années, l’apprentissage automatique supervisé a fait des progrès considérables dans l’identification des dommages aux bâtiments. Cependant, l'utilisation de l'apprentissage automatique supervisé reste difficile en raison des faits suivants : 1) les échantillons massifs de l'imagerie actuelle des dommages sont difficiles à étiqueter et ne peuvent donc pas satisfaire aux exigences de formation de l'apprentissage profond, et 2) la similarité entre les échantillons partiellement endommagés. et les bâtiments intacts sont élevés, ce qui empêche une classification précise. En tirant parti des nombreux échantillons de domaines auxiliaires, l'adaptation de domaine vise à transférer un classificateur formé par l'imagerie des dommages historiques à la tâche actuelle. Cependant, les approches traditionnelles d'adaptation de domaine ne prennent pas pleinement en compte les informations spécifiques à la catégorie lors de l'adaptation des fonctionnalités, ce qui pourrait entraîner un transfert négatif. Pour résoudre ce problème, nous proposons un nouveau cadre d'adaptation de domaine qui aligne individuellement chaque catégorie du domaine cible sur celle du domaine source. Notre méthode combine l'auto-encodeur variationnel (VAE) et le modèle de mélange gaussien (GMM). Premièrement, le GMM est établi pour caractériser la distribution du domaine source. Ensuite, le VAE est construit pour extraire la fonctionnalité du domaine cible. Enfin, la divergence Kullback-Leibler (KL) est minimisée pour forcer la fonctionnalité du domaine cible à observer le GMM du domaine source. Deux tâches de détection des dommages utilisant des images post-séisme et post-ouragan sont utilisées pour vérifier l'efficacité de notre méthode. Les expériences montrent que la méthode proposée obtient des améliorations de 4.4% et 9.5% respectivement par rapport à la méthode conventionnelle.
Daming LIN
Ministry of Transport
Jie WANG
North China University of Technology
Yundong LI
North China University of Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Daming LIN, Jie WANG, Yundong LI, "Unsupervised Building Damage Identification Using Post-Event Optical Imagery and Variational Autoencoder" in IEICE TRANSACTIONS on Information,
vol. E104-D, no. 10, pp. 1770-1774, October 2021, doi: 10.1587/transinf.2021EDL8034.
Abstract: Rapid building damage identification plays a vital role in rescue operations when disasters strike, especially when rescue resources are limited. In the past years, supervised machine learning has made considerable progress in building damage identification. However, the usage of supervised machine learning remains challenging due to the following facts: 1) the massive samples from the current damage imagery are difficult to be labeled and thus cannot satisfy the training requirement of deep learning, and 2) the similarity between partially damaged and undamaged buildings is high, hindering accurate classification. Leveraging the abundant samples of auxiliary domains, domain adaptation aims to transfer a classifier trained by historical damage imagery to the current task. However, traditional domain adaptation approaches do not fully consider the category-specific information during feature adaptation, which might cause negative transfer. To address this issue, we propose a novel domain adaptation framework that individually aligns each category of the target domain to that of the source domain. Our method combines the variational autoencoder (VAE) and the Gaussian mixture model (GMM). First, the GMM is established to characterize the distribution of the source domain. Then, the VAE is constructed to extract the feature of the target domain. Finally, the Kullback-Leibler (KL) divergence is minimized to force the feature of the target domain to observe the GMM of the source domain. Two damage detection tasks using post-earthquake and post-hurricane imageries are utilized to verify the effectiveness of our method. Experiments show that the proposed method obtains improvements of 4.4% and 9.5%, respectively, compared with the conventional method.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDL8034/_p
Copier
@ARTICLE{e104-d_10_1770,
author={Daming LIN, Jie WANG, Yundong LI, },
journal={IEICE TRANSACTIONS on Information},
title={Unsupervised Building Damage Identification Using Post-Event Optical Imagery and Variational Autoencoder},
year={2021},
volume={E104-D},
number={10},
pages={1770-1774},
abstract={Rapid building damage identification plays a vital role in rescue operations when disasters strike, especially when rescue resources are limited. In the past years, supervised machine learning has made considerable progress in building damage identification. However, the usage of supervised machine learning remains challenging due to the following facts: 1) the massive samples from the current damage imagery are difficult to be labeled and thus cannot satisfy the training requirement of deep learning, and 2) the similarity between partially damaged and undamaged buildings is high, hindering accurate classification. Leveraging the abundant samples of auxiliary domains, domain adaptation aims to transfer a classifier trained by historical damage imagery to the current task. However, traditional domain adaptation approaches do not fully consider the category-specific information during feature adaptation, which might cause negative transfer. To address this issue, we propose a novel domain adaptation framework that individually aligns each category of the target domain to that of the source domain. Our method combines the variational autoencoder (VAE) and the Gaussian mixture model (GMM). First, the GMM is established to characterize the distribution of the source domain. Then, the VAE is constructed to extract the feature of the target domain. Finally, the Kullback-Leibler (KL) divergence is minimized to force the feature of the target domain to observe the GMM of the source domain. Two damage detection tasks using post-earthquake and post-hurricane imageries are utilized to verify the effectiveness of our method. Experiments show that the proposed method obtains improvements of 4.4% and 9.5%, respectively, compared with the conventional method.},
keywords={},
doi={10.1587/transinf.2021EDL8034},
ISSN={1745-1361},
month={October},}
Copier
TY - JOUR
TI - Unsupervised Building Damage Identification Using Post-Event Optical Imagery and Variational Autoencoder
T2 - IEICE TRANSACTIONS on Information
SP - 1770
EP - 1774
AU - Daming LIN
AU - Jie WANG
AU - Yundong LI
PY - 2021
DO - 10.1587/transinf.2021EDL8034
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 10
JA - IEICE TRANSACTIONS on Information
Y1 - October 2021
AB - Rapid building damage identification plays a vital role in rescue operations when disasters strike, especially when rescue resources are limited. In the past years, supervised machine learning has made considerable progress in building damage identification. However, the usage of supervised machine learning remains challenging due to the following facts: 1) the massive samples from the current damage imagery are difficult to be labeled and thus cannot satisfy the training requirement of deep learning, and 2) the similarity between partially damaged and undamaged buildings is high, hindering accurate classification. Leveraging the abundant samples of auxiliary domains, domain adaptation aims to transfer a classifier trained by historical damage imagery to the current task. However, traditional domain adaptation approaches do not fully consider the category-specific information during feature adaptation, which might cause negative transfer. To address this issue, we propose a novel domain adaptation framework that individually aligns each category of the target domain to that of the source domain. Our method combines the variational autoencoder (VAE) and the Gaussian mixture model (GMM). First, the GMM is established to characterize the distribution of the source domain. Then, the VAE is constructed to extract the feature of the target domain. Finally, the Kullback-Leibler (KL) divergence is minimized to force the feature of the target domain to observe the GMM of the source domain. Two damage detection tasks using post-earthquake and post-hurricane imageries are utilized to verify the effectiveness of our method. Experiments show that the proposed method obtains improvements of 4.4% and 9.5%, respectively, compared with the conventional method.
ER -