Rapid post-disaster infrastructure damage characterisation using remote sensing and deep learning technologies: A tiered approach

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

Published: Jan. 5, 2025

Language: Английский

---

Globally scalable glacier mapping by deep learning matches expert delineation accuracy

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: Jan. 2, 2025

Abstract Accurate global glacier mapping is critical for understanding climate change impacts. Despite its importance, automated at a scale remains largely unexplored. Here we address this gap and propose Glacier-VisionTransformer-U-Net (GlaViTU), convolutional-transformer deep learning model, five strategies multitemporal global-scale using open satellite imagery. Assessing the spatial, temporal cross-sensor generalisation shows that our best strategy achieves intersection over union >0.85 on previously unobserved images in most cases, which drops to >0.75 debris-rich areas such as High-Mountain Asia increases >0.90 regions dominated by clean ice. A comparative validation against human expert uncertainties terms of area distance deviations underscores GlaViTU performance, approaching or matching expert-level delineation. Adding synthetic aperture radar data, namely, backscatter interferometric coherence, accuracy all where available. The calibrated confidence extents reported making predictions more reliable interpretable. We also release benchmark dataset covers 9% glaciers worldwide. Our results support efforts towards mapping.

Language: Английский

---

Improved Agricultural Field Segmentation in Satellite Imagery Using TL-ResUNet Architecture

Sensors, Journal Year: 2022, Volume and Issue: 22(24), P. 9784 - 9784

Published: Dec. 13, 2022

Currently, there is a growing population around the world, and this particularly true in developing countries, where food security becoming major problem. Therefore, agricultural land monitoring, use classification analysis, achieving high yields through efficient are important research topics precision agriculture. Deep learning-based algorithms for of satellite images provide more reliable accurate results than traditional algorithms. In study, we propose transfer learning based residual UNet architecture (TL-ResUNet) model, which semantic segmentation deep neural network model cover using images. The proposed combines strengths network, learning, architecture. We tested on public datasets such as DeepGlobe, showed that our outperforms classic models initiated with random weights pre-trained ImageNet coefficients. TL-ResUNet other several metrics commonly used accuracy performance measures tasks. Particularly, obtained an IoU score 0.81 validation subset DeepGlobe dataset model.

Language: Английский

---

Geospatial-based machine learning techniques for land use and land cover mapping using a high-resolution unmanned aerial vehicle image

Remote Sensing Applications Society and Environment, Journal Year: 2022, Volume and Issue: 29, P. 100859 - 100859

Published: Oct. 23, 2022

Language: Английский

---

Land Use and Land Cover Mapping with VHR and Multi-Temporal Sentinel-2 Imagery

Remote Sensing, Journal Year: 2023, Volume and Issue: 15(10), P. 2501 - 2501

Published: May 10, 2023

Land Use/Land Cover (LULC) mapping is the first step in monitoring urban sprawl and its environmental, economic societal impacts. While satellite imagery vegetation indices are commonly used for LULC mapping, limited resolution of these images can hamper object recognition Geographic Object-Based Image Analysis (GEOBIA). In this study, we utilize very high-resolution (VHR) optical with a 50 cm to improve GEOBIA classification. We focused on city Nice, France, identified ten classes using Random Forest classifier Google Earth Engine. investigate impact adding Gray-Level Co-Occurrence Matrix (GLCM) texture information spectral their temporal components, such as maximum value, standard deviation, phase amplitude from multi-spectral multi-temporal Sentinel-2 imagery. This work focuses identifying which input features result highest increase accuracy. The results show that single VHR image improves classification accuracy 62.62% 67.05%, especially when analysis not included. GLCM similar but smaller than image. Overall, inclusion 74.30%. blue band had largest classification, followed by green-red index normalized multi-band drought index.

Language: Английский

---

Utilizing Sentinel-2 Satellite Imagery for LULC and NDVI Change Dynamics for Gelephu, Bhutan

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(4), P. 1578 - 1578

Published: Feb. 16, 2024

Gelephu, located in the Himalayan region, has undergone significant development activities due to its suitable topography and geographic location. This led rapid urbanization recent years. Assessing land use cover (LULC) dynamics Normalized Difference Vegetation Index (NDVI) can provide important information about trends changes vegetation health, respectively. The of Geographic Information Systems (GIS) Remote Sensing (RS) techniques based on various satellite products offers a unique opportunity analyze these at local scale. Exploring Bhutan’s mandate maintain 60% forest analyzing LULC transitions using Sentinel-2 imagery 10 m resolution insights into potential future impacts. To examine these, we first performed mapping for Gelephu 2016 2023 Random Forest (RF) classifier identified changes. Second, study assessed change within area by analysing NDVI same period. Furthermore, also characterized resulting Thromde, sub-administrative municipal entity, as result notable intensity infrastructure activities. current used framework collect data, which was then pre-and post-processing create maps. classification model achieved high accuracy, with an under curve (AUC) up 0.89. corresponding statistics were analysed determine status indices, analysis reveals urban growth 5.65% 15.05% assessment shows deterioration health 75.11% loss healthy between 2023. results serve basis strategy adaption required environmental protection sustainable management, policy interventions minimize balance ecosystem, taking account landscape.

Language: Английский

---

Learning without Exact Guidance: Updating Large-Scale High-Resolution Land Cover Maps from Low-Resolution Historical Labels

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Journal Year: 2024, Volume and Issue: 64, P. 27717 - 27727

Published: June 16, 2024

Language: Английский

---

Machine Learning-Based Land Use and Land Cover Mapping Using Multi-Spectral Satellite Imagery: A Case Study in Egypt

Sustainability, Journal Year: 2023, Volume and Issue: 15(12), P. 9467 - 9467

Published: June 13, 2023

Satellite images provide continuous access to observations of the Earth, making environmental monitoring more convenient for certain applications, such as tracking changes in land use and cover (LULC). This paper is aimed develop a prediction model mapping LULC using multi-spectral satellite images, which were captured at spatial resolution 3 m by 4-band PlanetScope satellite. The dataset used study includes 105 geo-referenced categorized into 8 different classes. To train this on both raster vector data, various machine learning strategies Support Vector Machines (SVMs), Decision Trees (DTs), Random Forests (RFs), Normal Bayes (NB), Artificial Neural Networks (ANNs) employed. A set metrics including precision, recall, F-score, kappa index are utilized measure accuracy model. Empirical experiments conducted, results show that ANN achieved classification 97.1%. best our knowledge, represents first attempt monitor Egypt conducted high-resolution with resolution. highlights potential approach promoting sustainable practices contributing achievement development goals. proposed method can also reliable source improving geographical services, detecting changes.

Language: Английский

---

Land use classification in mine-agriculture compound area based on multi-feature random forest: a case study of Peixian

Frontiers in Sustainable Food Systems, Journal Year: 2024, Volume and Issue: 7

Published: Jan. 4, 2024

Introduction Land use classification plays a critical role in analyzing land use/cover change (LUCC). Remote sensing based on machine learning algorithm is one of the hot spots current remote technology research. The diversity surface objects and complexity their distribution mixed mining agricultural areas have brought challenges to traditional images, rich information contained images has not been fully utilized. Methods A quantitative difference index was proposed quantify select texture features easily confused types, random forest (RF) method with multi-feature combination schemes for developed, mine-agriculture compound area Peixian Xuzhou, China extracted. Results proved effective reducing dimensionality feature parameters resulted reduction optimal scheme dimension from 57 22. Among four methods scheme, RF emerged as most efficient accuracy 92.38% Kappa coefficient 0.90, which outperformed support vector (SVM), regression tree (CART), neural network (NN) algorithm. Conclusion findings indicate that differential novel approach discerning distinct among various types. It crucial selection optimization multispectral imagery. Random method, leveraging combination, provides fresh precise intricate ground within area.

Language: Английский

---

Machine learning versus deep learning in land system science: a decision-making framework for effective land classification

Frontiers in Remote Sensing, Journal Year: 2024, Volume and Issue: 5

Published: May 23, 2024

This review explores the comparative utility of machine learning (ML) and deep (DL) in land system science (LSS) classification tasks. Through a comprehensive assessment, study reveals that while DL techniques have emerged with transformative potential, their application LSS often faces challenges related to data availability, computational demands, model interpretability, overfitting. In many instances, traditional ML models currently present more effective solutions, as illustrated our decision-making framework. Integrative opportunities for enhancing accuracy include integration from diverse sources, development advanced architectures, leveraging unsupervised learning, infusing domain-specific knowledge. The research also emphasizes need regular evaluation, creation diversified training datasets, fostering interdisciplinary collaborations. Furthermore, promise future advancements is undeniable, considerations tip balance favor schemes. serves guide researchers, emphasizing importance choosing right tools evolving landscape LSS, achieve reliable nuanced land-use change data.

Language: Английский

---