Land Use and Land Cover Mapping Using Deep Learning Based Segmentation Approaches and VHR Worldview-3 Images

Remote Sensing, Journal Year: 2022, Volume and Issue: 14(18), P. 4558 - 4558

Published: Sept. 12, 2022

Deep learning-based segmentation of very high-resolution (VHR) satellite images is a significant task providing valuable information for various geospatial applications, specifically land use/land cover (LULC) mapping. The becomes more challenging with the increasing number and complexity LULC classes. In this research, we generated new benchmark dataset from VHR Worldview-3 twelve distinct classes two different geographical locations. We evaluated performance architectures encoders to find best design create highly accurate maps. Our results showed that DeepLabv3+ architecture an ResNeXt50 encoder achieved metric values IoU 89.46%, F-1 score 94.35%, precision 94.25%, recall 94.49%. This could be used by other researchers mapping similar or regions. Moreover, our can as reference implementing models via supervised, semi- weakly-supervised deep learning models. addition, model transfer generalizability methodologies.

Language: Английский

---

Enhancing Land Cover/Land Use (LCLU) classification through a comparative analysis of hyperparameters optimization approaches for deep neural network (DNN)

Ecological Informatics, Journal Year: 2023, Volume and Issue: 78, P. 102333 - 102333

Published: Oct. 11, 2023

Sustainable natural resources management relies on effective and timely assessment of conservation land practices. Using satellite imagery for Earth observation has become essential monitoring cover/land use (LCLU) changes identifying critical areas conserving biodiversity. Remote Sensing (RS) datasets are often quite large require tremendous computing power to process. The emergence cloud-based techniques presents a powerful avenue overcome limitations by allowing machine-learning algorithms process analyze RS the cloud. Our study aimed classify LCLU Talassemtane National Park (TNP) using Deep Neural Network (DNN) model incorporating five spectral indices differentiate six classes Sentinel-2 imagery. Optimization DNN was conducted comparative analysis three optimization algorithms: Random Search, Hyperband, Bayesian optimization. Results indicated that improved classification between with similar reflectance. Hyperband method had best performance, improving accuracy 12.5% achieving an overall 94.5% kappa coefficient 93.4%. dropout regularization prevented overfitting mitigated over-activation hidden nodes. initial results show machine learning (ML) applications can be tools management.

Language: Английский

---

Challenges for Sustainable Urban Planning: A Spatiotemporal Analysis of Complex Landslide Risk in a Latin American Megacity

Sustainability, Journal Year: 2024, Volume and Issue: 16(8), P. 3133 - 3133

Published: April 9, 2024

This paper analyzes the spatiotemporal evolution of a complex landslide risk scenario in Latin American megacity, underscoring key challenges it poses for sustainable urban planning such cities. research draws upon multiple studies commissioned by mayor’s office megacity Bogota, Colombia, and utilizes aerial photographs satellite imagery from diverse sensor types. The methodology used considered six analysis scenarios: rural/natural, mining, urban, risk, stabilization environmental park, informal reoccupation. findings reveal interplay between megacity’s peripheral areas, which face constraints human settlement, their potential construction material exploitation. relationship was further compounded weaknesses controlling occupations, coupled with burgeoning demand developable land context (landslide area: 73 ha). scenarios highlighted predominant use reactive approach that addressed events, changes, or problems after they had occurred, rather than proactively anticipating preventing risks at study site. detected land-use transformations unveiled different historical moments, culminating disaster (804 houses destroyed, 3000 families risk). catastrophe necessitated radical significant intervention, incurring substantial costs administration (USD 26.05 million). largest recorded one most extensive areas across America.

Language: Английский

---

Stylized facts of past 1000-year of China’s cropland changes

Land Use Policy, Journal Year: 2024, Volume and Issue: 144, P. 107258 - 107258

Published: July 1, 2024

Language: Английский

---

HexaLCSeg: A historical benchmark dataset from Hexagon satellite images for land cover segmentation [Software and Data Sets]

IEEE Geoscience and Remote Sensing Magazine, Journal Year: 2024, Volume and Issue: 12(3), P. 197 - 206

Published: Sept. 1, 2024

Language: Английский

---

Stakeholder-driven evaluation of the impact of land reform programme on landscape changes using earth observation data

Modeling Earth Systems and Environment, Journal Year: 2025, Volume and Issue: 11(2)

Published: Jan. 20, 2025

Language: Английский

---

Multitemporal Analysis of Declassified Keyhole Imagery’ for Landuse Change Detection in China (1960~1984): A Python-Based Spatial Coverage and Automation Workflow

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(5), P. 822 - 822

Published: Feb. 26, 2025

Keyhole imagery, acquired between the 1960s and 1980s, offers a unique opportunity to study land use changes prior era of modern remote sensing. This evaluates potential free-download imagery within China detect over five 5-year periods (1960–1984). Using metadata spatial analysis tools in Python 3.12, we classified images into three resolution categories (meter-level, five-meter-level, ten-meter-level) analyzed their distribution repeated coverage. Results show that 26.5%, 58.9%, 34.0% areas were capable detecting at least one land-use change event for respective categories. The T3 period (1970–1974) exhibited greatest diversity combinations among periods. However, uneven temporal coverage, particularly western rural regions, limits ability free conduct continuous multi-temporal analysis, collaboration with paid could fill gaps coverage improve accuracy detection. highlights historical research while underscoring need methodological refinements address data limitations. shared scripts processing techniques also support other using globally.

Language: Английский

---

Ensemble machine learning models for monitoring riparian vegetation dynamics using historical aerial orthophotos

Remote Sensing Applications Society and Environment, Journal Year: 2025, Volume and Issue: unknown, P. 101545 - 101545

Published: April 1, 2025

Language: Английский

---

Austrian Cadastre still in use – Example proceedings to determine the legal status of land property in southern Poland

Land Use Policy, Journal Year: 2023, Volume and Issue: 131, P. 106740 - 106740

Published: May 15, 2023

Language: Английский

---

Orthorectification of Large Datasets of Multi-scale Archival Aerial Imagery: A Case Study from Türkiye

Journal of Geovisualization and Spatial Analysis, Journal Year: 2023, Volume and Issue: 7(2)

Published: Aug. 8, 2023

Language: Английский

---