RT info:eu-repo/semantics/article
T1 Dilated residual U-Net with spatial–channel attention for building footprint extraction from UAV imagery in diverse landscapes
A1 Pham, Dung T.
A1 Tran, Duy X.
A1 Tran, Thuong V.
A1 Latorre Carmona, Pedro
A1 Bruce, David
A1 Zhu, Xuan
K1 Convolutional block attention module
K1 Transfer learning
K1 Semantic segmentation
K1 Urban and rural landscapes
K1 Automated mapping
K1 Teledetección
K1 Remote sensing
K1 Inteligencia artificial
K1 Artificial intelligence
AB Accurate extraction of building footprints from unmanned aerial vehicle (UAV) imageryis essential for urban planning, infrastructure monitoring, and disaster risk management.However, segmentation performance remains challenged by scale variability,spectral ambiguity, and partial occlusion in heterogeneous environments. Here, weexamine the integration of dilated convolution, residual learning, and spatial–channelattention within a unified U-Net framework for building footprint extraction. Ratherthan introducing new architectural components, the proposed approach investigateshow these established mechanisms influence segmentation behaviour across diverselandscape conditions. Multi-rate dilated convolution is incorporated in the bottleneckstage; residual connections support stable feature propagation, and attention modulesenhance feature discrimination. The model was evaluated on the WHU Building Datasetand a newly developed UAV dataset (HUMG) representing diverse environmentalconditions in Vietnam. Results showed improved segmentation performance relativeto U-Net, Feature Pyramid Network, and SegFormer, particularly in dense urban, rural,and vegetation-occluded environments. Although the proposed model incurs highercomputational costs, the additional complexity reflects both increased model capacityand integrated architectural design. The findings demonstrate that systematic adaptationof established mechanisms can enhance building segmentation robustness inhigh-resolution UAV imagery for offline geospatial applications.
PB Taylor & Francis
SN 1753-8947
YR 2026
FD 2026-08
LK https://hdl.handle.net/10259/12057
UL https://hdl.handle.net/10259/12057
LA eng
DS Repositorio Institucional de la Universidad de Burgos
RD 10-sep-2026