Fine-Scale Urban Informal Settlements Mapping by Fusing Remote Sensing Images and Building Data via a Transformer-Based Multimodal Fusion Network DOI
Runyu Fan, Fengpeng Li, Wei Han

et al.

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2022, Volume and Issue: 60, P. 1 - 16

Published: Jan. 1, 2022

Urban informal settlements (UIS) are high-density population with low standards of living and supply. UIS semantic segmentation, which identifies pixels corresponding to in remote sensing images, is crucial the estimation poor communities, urban management, resource allocation, future planning, particularly megacities. However, most studies on settlement mapping either based parcels (image classification) or (semantic segmentation). Few utilize object information improve mapping. Since formed by buildings (objects), utilizing can segmentation. Furthermore, current mainly focus using single-modality there a lack related research multimodal data. Due spatial heterogeneity settlements, only single modality image features limits effectiveness accuracy Aiming at achieving fine-scale results, this paper proposes segmentation method, namely UisNet, that utilizes transformer-based block receive data, including high-spatial-resolution images (parcel- pixel-level) building polygon data (object-level) identify UIS. The experiments were conducted Shenzhen City, they confirmed superior performance achieved an overall (OA) 94.80% mean intersection over union (mIoU) 85.51% testing set manually labeled dataset (UIS-Shenzhen dataset) outperformed best models tasks. Besides, we add public (GID compare our method state-of-the-art methods. Experiments show proposed UisNet improves mIoU 1.64% 7.58% compared other This work will be available https://github.com/RunyuFan/.

Language: Английский

Rapid post-disaster infrastructure damage characterisation using remote sensing and deep learning technologies: A tiered approach DOI Creative Commons
Nadiia Kopiika, Andreas Karavias, Pavlos Krassakis

et al.

Automation in Construction, Journal Year: 2025, Volume and Issue: 170, P. 105955 - 105955

Published: Jan. 5, 2025

Language: Английский

Citations

6
Assessing groundwater artificial recharge suitability in the Mi River basin using GIS, RS, and FAHP: a comprehensive analysis with seasonal variations DOI Creative Commons
Qilong Song, Yuyu Liu,

Zhongjie Wang

et al.

Applied Water Science, Journal Year: 2025, Volume and Issue: 15(2)

Published: Jan. 29, 2025

Language: Английский

Citations

2
3D building reconstruction from single street view images using deep learning DOI Creative Commons

Hui En Pang,

Filip Biljecki

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2022, Volume and Issue: 112, P. 102859 - 102859

Published: June 17, 2022

3D building models are an established instance of geospatial information in the built environment, but their acquisition remains complex and topical. Approaches to reconstruct often require existing (e.g. footprints) data such as point clouds, which scarce laborious acquire, limiting expansion. In parallel, street view imagery (SVI) has been gaining currency, driven by rapid expansion coverage advances computer vision (CV), it not used much for generating city models. Traditional approaches that can use SVI reconstruction multiple images, while practice, only few street-level images provide unobstructed a building. We develop from single image using image-to-mesh techniques modified CV domain. regard three scenarios: (1) standalone single-view reconstruction; (2) aided top delineating footprint; (3) refinement models, i.e. we examine enhance level detail block (LoD1) common. The results suggest trained supporting able overall geometry building, first scenario may derive approximate mass useful infer urban form cities. evaluate demonstrating usefulness volume estimation, with mean errors less than 10% last two scenarios. As is now available most countries worldwide, including many regions do have footprint and/or data, our method rapidly cost-effectively without requiring any information. Obtaining hitherto did any, enable number analyses locally time.

Language: Английский

Citations

69
A comprehensive framework for evaluating the quality of street view imagery DOI Creative Commons
Yujun Hou, Filip Biljecki

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2022, Volume and Issue: 115, P. 103094 - 103094

Published: Nov. 12, 2022

Language: Английский

Citations

59
Driving forces and prediction of urban land use change based on the geodetector and CA-Markov model: a case study of Zhengzhou, China DOI Creative Commons

Dehe Xu,

Ke Zhang,

Lianhai Cao

et al.

International Journal of Digital Earth, Journal Year: 2022, Volume and Issue: 15(1), P. 2246 - 2267

Published: Dec. 19, 2022

Exploring urban land use change is a classical problem in geography. Taking Zhengzhou as an example, this paper analyzes the spatial and temporal characteristics driving factors of change, simulates pattern future. The results study show that types city were mainly farmland construction land, area forestland, grassland, water area, unused was smaller, main transformation into land. accuracy check simulated type data 2020 showed kappa coefficient reached 0.9445, which met requirement. Then, according to predicted 2025, it found may have decreased, farmland, forestland increased. Based on force analysis changes, its prediction can provide important reference basis for formulation planning policies related construction.

Language: Английский

Citations

45
Classifying urban functional regions by integrating buildings and points-of-interest using a stacking ensemble method DOI Creative Commons
Min Yang, Bo Kong, Ruirong Dang

et al.

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2022, Volume and Issue: 108, P. 102753 - 102753

Published: March 29, 2022

The automatic classification of urban functional regions is vital for planning and governance. current methods mainly rely on single remote sensing image data or social data. However, these imagery-based have the disadvantage capturing high-level socioeconomic features, whereas information from alone rarely contains morphological features. To overcome limitations, it necessary to combine multisource functionalities. This study presents an ensemble method that combines vector-based buildings points-of-interest (POIs). For each block, we constructed improved graph convolutional neural network (GCNN) extract features constituent buildings. 'Word2Vec' model was used obtain characteristics POIs. On this basis, a stacking designed classifying functionality block. proposed trained tested in Nanshan District, Shenzhen, China. results showed accuracy 86.83%, which 12.2%–16.1% higher than standalone applications based single-source models were also applied two other districts, namely Futian Guangming, achieving accuracies 85.32% 68.37%, respectively, 3.68%–7.79% 3.69%–8.94% those obtained using single-sourced Moreover, improvements 2.41%–9.76%, compared with existing integration three areas. These suggest our can effectively integrate different sources provide alternative, higher-accuracy solution regions.

Language: Английский

Citations

42
Geographic mapping with unsupervised multi-modal representation learning from VHR images and POIs DOI
Lubin Bai, Weiming Huang, Xiuyuan Zhang

et al.

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2023, Volume and Issue: 201, P. 193 - 208

Published: June 1, 2023

Language: Английский

Citations

30
Multi-scale Feature Fusion and Transformer Network for urban green space segmentation from high-resolution remote sensing images DOI Creative Commons
Yong Cheng, Wei Wang, Zhoupeng Ren

et al.

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2023, Volume and Issue: 124, P. 103514 - 103514

Published: Oct. 5, 2023

Accurate extraction of urban green space is critical for preserving ecological balance and enhancing life quality. However, due to the complex morphology (e.g., different sizes shapes), it still challenging extract effectively from high-resolution image. To address this issue, we proposed a novel hybrid method, Multi-scale Feature Fusion Transformer Network (MFFTNet), as new deep learning approach extracting (GF-2) Our method was characterized by two aspects: (1) multi-scale feature fusion module transformer network that enhanced recovery edge information (2) vegetation (NDVI) highlighted boundaries identification. The GF-2 image utilized build labeled datasets, namely Greenfield Greenfield2. We compared MFFTNet with existing popular models (like PSPNet, DensASPP, etc.) evaluate effectiveness Mean Intersection Over Union (MIOU) benchmark on Greenfield, Greenfield2, public dataset (WHDLD). Experiments Greenfield2 showed can achieve high MIOU (86.50%), which outperformed networks like PSPNet DensASPP 0.86% 3.28%, respectively. Meanwhile, incorporating further achieved 86.76% experimental results demonstrate outperforms state-of-the-art methods in segmentation.

Language: Английский

Citations

23
ELGC-Net: Efficient Local–Global Context Aggregation for Remote Sensing Change Detection DOI

Mubashir Noman,

Mustansar Fiaz, Hisham Cholakkal

et al.

IEEE Transactions on Geoscience and Remote Sensing, Journal Year: 2024, Volume and Issue: 62, P. 1 - 11

Published: Jan. 1, 2024

Deep learning has shown remarkable success in remote sensing change detection (CD), aiming to identify semantic regions between co-registered satellite image pairs acquired at distinct time stamps. However, existing convolutional neural network (CNN) and transformer-based frameworks often struggle accurately segment regions. Moreover, transformers-based methods with standard self-attention suffer from quadratic computational complexity respect the resolution, making them less practical for CD tasks limited training data. To address these issues, we propose an efficient framework, ELGC-Net, which leverages rich contextual information precisely estimate while reducing model size. Our ELGC-Net comprises a Siamese encoder, fusion modules, decoder. The focus of our design is introduction Efficient Local-Global Context Aggregator (ELGCA) module within capturing enhanced global context local spatial through novel pooled-transpose (PT) attention depthwise convolution, respectively. PT employs pooling operations robust feature extraction minimizes cost transposed attention. Extensive experiments on three challenging datasets demonstrate that outperforms methods. Compared recent approach (ChangeFormer), achieves 1.4% gain intersection over union (IoU) metric LEVIR-CD dataset, significantly trainable parameters. proposed sets new state-of-the-art performance benchmarks. Finally, also introduce ELGC-Net-LW, lighter variant reduced complexity, suitable resource-constrained settings, achieving comparable performance. source code publicly available https://github.com/techmn/elgcnet.

Language: Английский

Citations

16
An MIU-based deep embedded clustering model for urban functional zoning from remote sensing images and VGI data DOI Creative Commons
Anqi Lin, Bo Huang, Hao Wu

et al.

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 128, P. 103689 - 103689

Published: Feb. 21, 2024

Urban functional zoning offers valuable insights into urban morphology and sustainable development. However, the conventional fixed spatial units, such as blocks grids, cannot easily capture morphological characteristics inherent in union separation during evolution. In this paper, by taking advantage of remote sensing images geospatial big data, we propose a minimum identification unit (MIU)-based model. This approach integrates deep embedded clustering buildings to generate segmentation, then identifies function generating semantic vectors with Word2Vec The effectiveness proposed method was tested city Wuhan China. results highlight that MIUs provide more flexible suitable for segmenting zones compared traditional street blocks. is feasible way deal redundancy volunteered geographic information (VGI) data when identifying function, quality issue only has significant impact on minor types. Moreover, building can effectively reveal fine-scale structure, especially administration, manufacturing, residential demonstrates potential our enhancing understanding supporting

Language: Английский

Citations

15