DS Net: A Dual-Coded Segmentation Network Leveraging Large Model Prior Knowledge for Intelligent Landslide Extraction DOI Creative Commons
Xiao Wang,

Dongsheng Zhong,

Chenghao Liu

и другие.

Remote Sensing, Год журнала: 2025, Номер 17(11), С. 1912 - 1912

Опубликована: Май 31, 2025

Landslides are characterized by their suddenness and destructive power, making rapid accurate identification crucial for emergency rescue disaster assessment in affected areas. To address the challenges of limited landslide samples data complexity, a sample library was constructed using high-resolution remote sensing imagery combined with field validation. An innovative Dual-Coded Segmentation Network (DS Net), which realizes dynamic alignment deep fusion local details global context, image features domain knowledge through multi-attention mechanism Prior Knowledge Integration (PKI) module Cross-Feature Aggregation (CFA) module, significantly improves detection accuracy reliability. objectively evaluate performance DS Net model, four efficient semantic segmentation models—SegFormer, SegNeXt, FeedFormer, U-MixFormer—were selected comparison. The results demonstrate that achieves superior (overall = 0.926, precision 0.884, recall 0.879, F1-score 0.882), metrics 3.5–7.1% higher than other models. These findings confirm effectively efficiency identification, providing critical scientific basis prevention mitigation.

Язык: Английский

Convolutional Neural Network-Based Risk Assessment of Regional Susceptibility to Road Collapse Disasters: A Case Study in Guangxi DOI Creative Commons

Cheng Li,

Zhixiang Lu,

Yulong Hu

и другие.

Applied Sciences, Год журнала: 2025, Номер 15(6), С. 3108 - 3108

Опубликована: Март 13, 2025

The Guangxi Zhuang Autonomous Region, a vital strategic geographic entity in southern China, is prone to frequent road collapse disasters due its complex topography and high rainfall, severely affecting regional economic social development. Existing risk assessments for these often lack comprehensive analysis of the combined influence multiple factors, their predictive accuracy requires enhancement. To address deficiencies, this study utilized ResNet18 model, convolutional neural network (CNN)-based approach, integrating 10 critical factors—including slope gradient, lithology, precipitation—to develop assessment model disasters. This predicts maps spatial distribution across Guangxi. results reveal that very high-risk areas span 49,218.94 km2, constituting 20.38% Guangxi’s total area, with disaster point density 8.67 per 100 km2; cover 56,543.87 representing 23.41%, 3.38 low-risk encompass 61,750.69 accounting 25.57%, 0.29 km2. receiver operating characteristic (ROC) curve yields an area under (AUC) value 0.7879, confirming model’s reliability assessing risk. establishes scientific foundation prevention mitigation offers valuable guidance similar regions.

Язык: Английский

Процитировано

0
Debris-flow susceptibility assessment using deep learning algorithms with GeoDetector for factor optimization DOI
Kun Li, Junsan Zhao, Guoping Chen

и другие.

Bulletin of Engineering Geology and the Environment, Год журнала: 2025, Номер 84(6)

Опубликована: Май 7, 2025

Язык: Английский

Процитировано

0
DS Net: A Dual-Coded Segmentation Network Leveraging Large Model Prior Knowledge for Intelligent Landslide Extraction DOI Creative Commons
Xiao Wang,

Dongsheng Zhong,

Chenghao Liu

и другие.

Remote Sensing, Год журнала: 2025, Номер 17(11), С. 1912 - 1912

Опубликована: Май 31, 2025

Landslides are characterized by their suddenness and destructive power, making rapid accurate identification crucial for emergency rescue disaster assessment in affected areas. To address the challenges of limited landslide samples data complexity, a sample library was constructed using high-resolution remote sensing imagery combined with field validation. An innovative Dual-Coded Segmentation Network (DS Net), which realizes dynamic alignment deep fusion local details global context, image features domain knowledge through multi-attention mechanism Prior Knowledge Integration (PKI) module Cross-Feature Aggregation (CFA) module, significantly improves detection accuracy reliability. objectively evaluate performance DS Net model, four efficient semantic segmentation models—SegFormer, SegNeXt, FeedFormer, U-MixFormer—were selected comparison. The results demonstrate that achieves superior (overall = 0.926, precision 0.884, recall 0.879, F1-score 0.882), metrics 3.5–7.1% higher than other models. These findings confirm effectively efficiency identification, providing critical scientific basis prevention mitigation.

Язык: Английский

Процитировано

0