Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment

Yu, Jiaxi; Fukuda, Tomohiro; Yabuki, Nobuyoshi

doi:https://doi.org/10.1016/j.ijdrr.2025.105664

Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment

Yu, Jiaxi; Fukuda, Tomohiro; Yabuki, Nobuyoshi

International Journal of Disaster Risk Reduction, 2025, 128, 105664

Number of Access:22（2025-11-23 22:38 Counts）

Identifier to cite or link to this item: https://hdl.handle.net/11094/102780

Link to primary information

File	Format	Terms of use	Size	Views	Date.Available	Description	information
IntJDisasterRiskReduct_128	pdf	None	4.05 MB	30	2025.09.26

Creative Commons : Attribution

Item Information

Output File Export EndNote Basic Export Mendeley

Title	Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment
Creator	CreatorName Yu, Jiaxi
	NRID 1000080379114 1000080379114 ORCID 0000-0002-4271-4445 0000-0002-4271-4445 researchmap permalink tomohirofukuda tomohirofukuda Scopus Author ID 57211973503 57211973503 CreatorName Fukuda, Tomohiro
	NRID 1000050312434 1000050312434 CreatorName Yabuki, Nobuyoshi
Subject	Attention mechanisms
	Benchmark performance
	Consistency regularization
	Flood disaster
	Prior knowledge
	Semi-supervised learning
Description	Yu J., Fukuda T., Yabuki N.. Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment. International Journal of Disaster Risk Reduction 128, 105664 (2025); https://doi.org/10.1016/j.ijdrr.2025.105664.
Abstract	Rapid and accurate building damage assessment (BDA) following floods is critical for effective disaster response, yet faces challenges from limited labeled data and subtle damage cues in satellite imagery. Existing deep learning change detection (CD) methods may exhibit low recall or misclassify damage inappropriately when transferred directly to the post-flood BDA (Flood-BDA) task. This study addresses these gaps by establishing the first systematic benchmark evaluating both supervised CD model transfer and semi-supervised learning (SSL) specifically for Flood-BDA. This research investigate image-level consistency regularization SSL to combat data scarcity, finding that strategies using pseudo-label derived reference distributions significantly enhance performance (+1.17% avg. Kappa at 5% labels). Notably, pseudo-label outperform ground-truth label strategies in low-label settings (e.g., +4.84% Kappa at 5% labels). Furthermore, confronting the limitations of transferred CD models (low recall, misclassifying ’destroyed’ as ’no damage’), this paper proposed a simple prior attention disaster assessment Net (SPADANet), a lightweight U-Net incorporating a simple prior attention module designed for Flood-BDA. SPADANet demonstrably improves recall (+9.22% over best CD baseline) and exhibits more favorable error patterns for Flood-BDA, despite a precision trade-off. This work provides crucial benchmarks, validates the need for recall-driven, DA-specific designs distinct from CD, and demonstrates the potential of prior attention and image-level consistency regularization for post-flood building damage assessment. The code will be available at https://github.com/JX-OctoNeko/Flood_BDA_benchmark.git
Publisher	Elsevier Ltd
Source Title	International Journal of Disaster Risk Reduction
Volume (Issue)	128
Page	105664
Date of Issued	2025-10-01
Language	English
Handle URL	https://hdl.handle.net/11094/102780
PISSN	22124209
Relation.isIdenticalTo	DOI (Publisher Version) https://doi.org/10.1016/j.ijdrr.2025.105664
Access Rights	open access
CopyRight	This article is licensed under a Creative Commons Attribution 4.0 International License.
CopyRight	© 2025 The Authors
oaire:version	Version of Record
Category	学術雑誌論文 Journal Article

Resource Type	journal article
Local Type	Journal Article
DCMI Type	Text
DCTERMS.bibliographicCitation	International Journal of Disaster Risk Reduction. 2025, 128, p. 105664
DC.title	Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment
DC.creator	Yu, Jiaxi
	Fukuda, Tomohiro
	Yabuki, Nobuyoshi
DC.publisher	Elsevier Ltd
DC.language' scheme='DCTERMS.RFC1766	English
DC.type' scheme='DCTERMS.DCMIType	Text
DCTERMS.issued' scheme='DCTERMS.W3CDTF	2025-10-01
DC.identifier	https://doi.org/10.1016/j.ijdrr.2025.105664
DC.subject	Attention mechanisms
	Benchmark performance
	Consistency regularization
	Flood disaster
	Prior knowledge
	Semi-supervised learning
DC.description	Yu J., Fukuda T., Yabuki N.. Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment. International Journal of Disaster Risk Reduction 128, 105664 (2025); https://doi.org/10.1016/j.ijdrr.2025.105664.
DCTERMS.abstract	Rapid and accurate building damage assessment (BDA) following floods is critical for effective disaster response, yet faces challenges from limited labeled data and subtle damage cues in satellite imagery. Existing deep learning change detection (CD) methods may exhibit low recall or misclassify damage inappropriately when transferred directly to the post-flood BDA (Flood-BDA) task. This study addresses these gaps by establishing the first systematic benchmark evaluating both supervised CD model transfer and semi-supervised learning (SSL) specifically for Flood-BDA. This research investigate image-level consistency regularization SSL to combat data scarcity, finding that strategies using pseudo-label derived reference distributions significantly enhance performance (+1.17% avg. Kappa at 5% labels). Notably, pseudo-label outperform ground-truth label strategies in low-label settings (e.g., +4.84% Kappa at 5% labels). Furthermore, confronting the limitations of transferred CD models (low recall, misclassifying ’destroyed’ as ’no damage’), this paper proposed a simple prior attention disaster assessment Net (SPADANet), a lightweight U-Net incorporating a simple prior attention module designed for Flood-BDA. SPADANet demonstrably improves recall (+9.22% over best CD baseline) and exhibits more favorable error patterns for Flood-BDA, despite a precision trade-off. This work provides crucial benchmarks, validates the need for recall-driven, DA-specific designs distinct from CD, and demonstrates the potential of prior attention and image-level consistency regularization for post-flood building damage assessment. The code will be available at https://github.com/JX-OctoNeko/Flood_BDA_benchmark.git
DC.rights	This article is licensed under a Creative Commons Attribution 4.0 International License.
DC.rights	© 2025 The Authors
DC.format	102780
citation_title	Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment
citation_author	Yu, Jiaxi
	Fukuda, Tomohiro
	Yabuki, Nobuyoshi
citation_publisher	Elsevier Ltd
citation_language	English
citation_date	2025-10-01
citation_journal_title	International Journal of Disaster Risk Reduction
citation_volume	128
citation_firstpage	105664
citation_doi	https://doi.org/10.1016/j.ijdrr.2025.105664
citation_public_url	https://hdl.handle.net/11094/102780

Benchmarking attention mechanisms and consistency regularization semi-supervised learning for post-flood building damage assessment

Link to primary information

Item Information

Show details