DBEFNet: a dual-branch fusion network integrating remote sensing imagery and building morphology for urban village segmentation

Urban villages (UVs) are informal settlements embedded within Chinese cities, and their boundary delineation is fundamental to urban renewal and refined spatial management. However, accurate segmentation of UVs relies on both visual characteristics and the spatial organization of buildings, yet previous studies have often relied on simplified representations of building morphology that inadequately capture spatial configuration. We propose DBEFNet, a dual-branch encoding and fusion framework that exploits remote sensing imagery (RSI) and building morphology heatmaps for fine-grained UVs segmentation. The RSI branch combines CNN and Swin Transformer encoders with multi-scale feature extraction to capture and integrate local appearance and broader spatial context. In the Building branch, multiple building attributes are transformed into morphology heatmaps and adaptively weighted to characterize building-cluster patterns. A hybrid attention fusion module (HAFM) models bidirectional dependencies between visual semantics and building morphology. Experiments across 15 cities yielded an overall accuracy (OA) of 94.79 % and a mean Intersection over Union (mIoU) of 86.24 %. Ablation experiments indicate that incorporating building morphology improves mIoU by 0.87 % over the enhanced RSI-only model, while adding the HAFM provides further gains of 0.93 %. Compared with the RSI-only model, DBEFNet achieved higher mIoU in 11 of the 15 cities, with an average city-wise gain of 1.92 %. Trained on Beijing, DBEFNet achieved 75.79 % mean mIoU across 12 unseen target cities, demonstrating its cross-city generalization capability. These results underscore the critical role of building morphology and the value of integrating morphological cues with visual appearance for precise UVs mapping.

Authors

Institutions

Publication Details

Journal
International Journal of Applied Earth Observation and Geoinformation
Published
2026-11-14
DOI
https://doi.org/10.1016/j.jag.2026.105682
Primary Topic
Automated Road and Building Extraction
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

DBEFNet: a dual-branch fusion network integrating remote sensing imagery and building morphology for urban village segmentation

Yimin Chen, Wufan Zhao, Haifeng Li, Zhonglin Yang et al.
International Journal of Applied Earth Observation and Geoinformation
Automated Road and Building Extraction
article

DBEFNet: a dual-branch fusion network integrating remote sensing imagery and building morphology for urban village segmentation

Yimin Chen, Wufan Zhao, Haifeng Li, Zhonglin Yang, Mengwei Liu, Ying Sun
article en

Abstract

Urban villages (UVs) are informal settlements embedded within Chinese cities, and their boundary delineation is fundamental to urban renewal and refined spatial management. However, accurate segmentation of UVs relies on both visual characteristics and the spatial organization of buildings, yet previous studies have often relied on simplified representations of building morphology that inadequately capture spatial configuration. We propose DBEFNet, a dual-branch encoding and fusion framework that exploits remote sensing imagery (RSI) and building morphology heatmaps for fine-grained UVs segmentation. The RSI branch combines CNN and Swin Transformer encoders with multi-scale feature extraction to capture and integrate local appearance and broader spatial context. In the Building branch, multiple building attributes are transformed into morphology heatmaps and adaptively weighted to characterize building-cluster patterns. A hybrid attention fusion module (HAFM) models bidirectional dependencies between visual semantics and building morphology. Experiments across 15 cities yielded an overall accuracy (OA) of 94.79 % and a mean Intersection over Union (mIoU) of 86.24 %. Ablation experiments indicate that incorporating building morphology improves mIoU by 0.87 % over the enhanced RSI-only model, while adding the HAFM provides further gains of 0.93 %. Compared with the RSI-only model, DBEFNet achieved higher mIoU in 11 of the 15 cities, with an average city-wise gain of 1.92 %. Trained on Beijing, DBEFNet achieved 75.79 % mean mIoU across 12 unseen target cities, demonstrating its cross-city generalization capability. These results underscore the critical role of building morphology and the value of integrating morphological cues with visual appearance for precise UVs mapping.

International Journal of Applied Earth Observation and GeoinformationVol. 155
Central South University (CN), Sun Yat-sen University (CN), Guangxi Academy of Sciences (CN), The Hong Kong University of Science and Technology (Guangzhou) (CN)
National Natural Science Foundation of China
Sustainable cities and communities
Automated Road and Building Extraction
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon