RT info:eu-repo/semantics/article
T1 SAGRNet: A novel object-based graph convolutional neural network for diverse vegetation cover classification in remotely-sensed imagery
A1 Gui, Baoling
A1 Sam, Lydia
A1 Bhardwaj, Anshuman
A1 Soto Gómez, Diego
A1 González Peñaloza, Félix
A1 Buchroithner, Manfred F.
A1 Green, David R.
K1 Object-based classification
K1 Graph convolutional
K1 Vegetation mapping
K1 Deep learning
K1 Remote sensing
K1 Aprendizaje automático
K1 Machine learning
K1 Cartografía de la vegetación
K1 Vegetation mapping
AB Growing global population, changing climate, and shrinking land resources demand for quicker, efficient, andmore accurate methods of mapping and monitoring vegetation cover in remote sensing datasets. Many deeplearning-based methods have been widely applied for semantic segmentation tasks in remote sensing images ofvegetated environments. However, most existing models are pixel-based, which introduces challenges such ashigh time consumption, cumbersome implementation, and limited scalability. This paper presents the SAGRNetmodel, a Graph Convolutional Neural Network (GCN) that incorporates sampling aggregation and self-attentionmechanisms, while leveraging the ResNet residual network structure. A key innovation of SAGRNet is its abilityto fuse features extracted through diverse algorithms, enabling comprehensive representation and enhancedclassification performance. The SAGRNet model demonstrates superior performance over leading pixel-basedneural networks, such as U-Net++ and DeepLabV3, in terms of both time efficiency and accuracy in vegetation image classification tasks. We achieved an overall mapping accuracy of ~90 % using SAGRNet, compared to~87% and ~85% from U-Net++ and DeepLabV3, respectively. Additionally, it offers more convenience in dataprocessing. Furthermore, the model significantly outperforms cutting-edge graph-based convolutional networks,including Graph U-Net (achieved overall accuracy ~65%) and TGNN (achieved overall accuracy ~75%),showcasing exceptional generalization capability and classification accuracy. This paper provides a comprehensive analysis of the various processing aspects of this object-based GCN for vegetation mapping and emphasizes its significant potential for practical use. The model’s versatility can also be expanded to other imageprocessing domains, offering unprecedented possibilities of information extraction from satellite imagery
PB Elsevier
SN 0924-2716
YR 2025
FD 2025-09
LK https://hdl.handle.net/10259/11723
UL https://hdl.handle.net/10259/11723
LA eng
NO The authors acknowledge BBSRC International Partnership Award (Ref: RG17324-19) and the University of Aberdeen to fund the research activities. We would like to thank the UK Centre for Ecology and Hydrology (UKCEH) for providing the high-resolution land cover maps of the United Kingdom, which served as an essential reference in our regional experiments. We also express our appreciation to ESRI for releasing the Global Land Cover dataset through the Living Atlas platform, which enabled the global-scale validation of our model. These openly accessible and high-quality datasets significantly supported the development and evaluation of our work
DS Repositorio Institucional de la Universidad de Burgos
RD 26-may-2026