Advances in Deep Learning Recognition of Landslides Based on Remote Sensing Images

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(10), P. 1787 - 1787

Published: May 18, 2024

Against the backdrop of global warming and increased rainfall, hazards potential risks landslides are increasing. The rapid generation a landslide inventory is great significance for disaster prevention reduction. Deep learning has been widely applied in identification due to its advantages terms deeper model structure, high efficiency, accuracy. This article first provides an overview deep technology basic principles, as well current status remote sensing databases. Then, classic recognition models such AlexNet, ResNet, YOLO, Mask R-CNN, U-Net, Transformer, EfficientNet, DeeplabV3+ PSPNet were introduced, limitations each extensively analyzed. Finally, constraints summarized, development direction was purpose this promote in-depth research order provide academic references mitigation disasters post-disaster rescue work. results indicate that methods have characteristics efficiency accuracy automatic recognition, more attention should be paid emerging future.

Language: Английский

---

Landslide mapping based on a hybrid CNN-transformer network and deep transfer learning using remote sensing images with topographic and spectral features

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2023, Volume and Issue: 126, P. 103612 - 103612

Published: Dec. 16, 2023

Landslides frequently cause serious property damage and casualties. Therefore, it is crucial to have rapid accurate landslide mapping (LM) support post-earthquake assessment emergency rescue efforts. Many studies been conducted in recent years on the application of automatic LM methods using remote sensing images (RSIs). However, existing face challenges accurately distinguishing landslides due problems large differences features scales among landslides, as well similarities different ground objects optical RSIs. Here, we propose a semantic segmentation model called SCDUNet++, which combines advantages convolutional neural network (CNN) transformer enhance discrimination extraction features. Then, constructed multi-channel dataset Luding Jiuzhaigou earthquake areas Sentinel-2 NASADEM data. We evaluated performance SCDUNet++ this dataset. The results showed that can extract fuse spectral topographic information more effectively. Compared with other state-of-the-art models, achieved highest IoU F1 score all four test areas. In addition, models significant improvements area after knowledge transfer fine-tuning. direct prediction, eight namely DeepLabv3+, Segformer, TransUNet, SwinUNet, STUNet, UNet, UNet++, demonstrated ranging from 8.33% 27.5% 6.58% 23.67% implementing deep learning (DTL). This finding highlights practicality DTL for cross-domain data-poor

Language: Английский

---

Landslide Identification from Post-Earthquake High-Resolution Remote Sensing Images Based on ResUNet–BFA

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(6), P. 995 - 995

Published: March 12, 2025

The integration of deep learning and remote sensing for the rapid detection landslides from high-resolution imagery plays a crucial role in post-disaster emergency response. However, availability publicly accessible datasets specifically landslide remains limited, posing challenges researchers meeting task requirements. To address this issue, study develops releases dataset using Google Earth imagery, focusing on impact zones 2008 Wenchuan Ms8.0 earthquake, 2014 Ludian Ms6.5 2017 Jiuzhaigou Ms7.0 earthquake as research areas. contains 2727 samples with spatial resolution 1.06 m. enhance recognition, lightweight boundary-focused attention (BFA) mechanism designed Canny operator is adopted. This improves model’s ability to emphasize edge features integrated ResUNet model, forming ResUNet–BFA architecture identification. experimental results indicate that model outperforms widely used algorithms extracting boundaries details, resulting fewer misclassifications omissions. Additionally, compared conventional mechanisms, BFA achieves superior performance, producing recognition more closely align actual labels.

Language: Английский

---

Landslide Detection in Satellite Images using InceptionU-Net and Convolutional Block Attention Module

Procedia Computer Science, Journal Year: 2025, Volume and Issue: 258, P. 4301 - 4310

Published: Jan. 1, 2025

Language: Английский

---

GDSNet: A gated dual-stream convolutional neural network for automatic recognition of coseismic landslides

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 127, P. 103677 - 103677

Published: Feb. 1, 2024

Automatic recognition of numerous coseismic landslides after a violent earthquake is crucial for emergency rescue and post-disaster reconstruction. Currently, deep learning techniques have achieved state-of-the-art performance in landslide recognition. However, Convolutional Neural Networks (CNNs) often lose detailed information during downsampling cannot adequately learn changeable shapes, colors, sizes landslides. In addition, complicated backgrounds, e.g., bare slopes dry riverbeds, are easily misidentified as Focusing on the above difficulties, this work proposes Gated Dual-Stream Network (GDSNet) recognition, which contains two branches feature fusion module. One branch, called CPSConv, can extract sizes, spectra, textures guarantee accurate identification small Another branch utilizes gated convolution strategy to adjust weights importance enhance features suppress background features. The aggregation module fuses from effectively improve accuracy GDSNet applied four earthquakes by model training testing. test dataset, compared with 9 models, mIoU, F1, Kappa coefficient values improved at least 12.30%, 8.12%, 16.25%, respectively. 1.08%-37.68% than other models.

Language: Английский

---

Enhancing deep learning-based landslide detection from open satellite imagery via multisource data fusion of spectral, textural, and topographical features: a case study of old landslide detection in the Three Gorges Reservoir Area (TGRA)

Geocarto International, Journal Year: 2024, Volume and Issue: 39(1)

Published: Jan. 1, 2024

Language: Английский

---

Review on the Artificial Intelligence-based methods in Landslide Detection and Susceptibility Assessment: Current Progress and Future Directions

Intelligent geoengineering., Journal Year: 2024, Volume and Issue: 1(1), P. 1 - 18

Published: Nov. 2, 2024

Language: Английский

---

Versatility Evaluation of Landslide Risk with Window Sizes and Sampling Techniques Based on Deep Learning

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(22), P. 10571 - 10571

Published: Nov. 16, 2024

Across the globe, landslides cause significant loss of life, injuries, and widespread damage to homes infrastructure. Therefore, assessing analyzing landslide hazards is crucial human, environmental, cultural, economic, social sustainability. This study utilizes ArcGIS 10.8 Python 3.9 create databases for Niigata Prefecture (NIG), Iwate Miyagi Prefectures (IWT-MYG), Hokkaido (HKD), drawing on data obtained from National Research Institute Earth Science Disaster Resilience, Japan. A distinguishing feature this application a Convolutional Neural Network (CNN), which significantly outperforms traditional machine learning models in image-based pattern recognition by extracting contextual information surrounding areas, distinct advantage image tasks. Unlike conventional methods that often require manual selection engineering, CNNs automate extraction, enabling more nuanced understanding complex patterns. By experimenting with CNN input window sizes ranging 3 × 27 pixels employing diverse sampling techniques, we demonstrate larger windows enhance model’s predictive accuracy capturing wider range environmental interactions critical effective modeling. 19 pixel typically yield best overall performance, CNN-19 achieving an AUC 0.950, 0.982 0.969 NIG, HKD, IWT-MYG, respectively. Furthermore, improve prediction reliability using oversampling random window-moving method. For instance, NIG region, 0.983, while downsampling 0.950). These less commonly applied approaches detection, help address issue imbalance seen datasets, where instances are far outnumbered non-landslide occurrences. While challenges remain enhancing generalization, research makes progress developing robust adaptable tools prediction, vital ensuring societal resilience.

Language: Английский

---

LandslideNet: A landslide semantic segmentation network based on single-temporal optical remote sensing images

Advances in Space Research, Journal Year: 2024, Volume and Issue: 74(10), P. 4616 - 4638

Published: July 17, 2024

Language: Английский

---

Geological remote sensing interpretation via a local-to-global sensitive feature fusion network

International Journal of Applied Earth Observation and Geoinformation, Journal Year: 2024, Volume and Issue: 135, P. 104258 - 104258

Published: Nov. 22, 2024

Language: Английский

---