Scholarlink

<p>Addressing the challenges of difficult deformation signal detection and low spatial positioning accuracy in identifying potential landslide hazards in mountainous regions, this paper proposes a multi-tiered and multi-model collaborative identification approach. Firstly, surface deformation time-series data were acquired using SBAS-InSAR technology. Anomalous deformation areas are identified through spatial autocorrelation analysis (Local Moran’s I), and these deformation data are spatially constrained by integrating susceptibility evaluation results generated from a Random Forest model and the Information Value method. This enabled preliminary screening of potential hazard zones via kernel density estimation. Subsequently, a U-Net model was employed to perform semantic segmentation on Sentinel-2 imagery. A secondary identification model, constructed based on geomechanical evolutionary principles, further refines the initial screening results to ultimately determine the distribution of potential landslide areas.Field validation results demonstrate that the collaborative multi-model identification outcomes are highly consistent with GNSS monitoring data in terms of deformation trends, maintaining an error confidence interval within 10&#xa0;mm. The identified potential landslides show high spatial correlations: an 84.6% spatial concordance with Quaternary cover layers, an 81.8% spatial concordance with dip slope structures, and a strong alignment with historical landslide distributions. Furthermore, the U-Net model achieved an AUC of 0.88 on the test set and an AUC of 0.81 on the validation set. This study effectively enhances the accuracy and reliability of landslide hazard identification by synergizing three modules—InSAR time-series analysis, machine learning classification, and deep learning segmentation—thereby providing a novel technical pathway for landslide disaster prevention in mountainous regions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-model landslide integrated hazard detection using SBAS-InSAR and U-Net segmentation

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Xinyu Cheng,
  • Bin Zeng,
  • Luyao Tang,
  • Jingjing Yuan,
  • Dong Ai,
  • Wei Huang

摘要

Addressing the challenges of difficult deformation signal detection and low spatial positioning accuracy in identifying potential landslide hazards in mountainous regions, this paper proposes a multi-tiered and multi-model collaborative identification approach. Firstly, surface deformation time-series data were acquired using SBAS-InSAR technology. Anomalous deformation areas are identified through spatial autocorrelation analysis (Local Moran’s I), and these deformation data are spatially constrained by integrating susceptibility evaluation results generated from a Random Forest model and the Information Value method. This enabled preliminary screening of potential hazard zones via kernel density estimation. Subsequently, a U-Net model was employed to perform semantic segmentation on Sentinel-2 imagery. A secondary identification model, constructed based on geomechanical evolutionary principles, further refines the initial screening results to ultimately determine the distribution of potential landslide areas.Field validation results demonstrate that the collaborative multi-model identification outcomes are highly consistent with GNSS monitoring data in terms of deformation trends, maintaining an error confidence interval within 10 mm. The identified potential landslides show high spatial correlations: an 84.6% spatial concordance with Quaternary cover layers, an 81.8% spatial concordance with dip slope structures, and a strong alignment with historical landslide distributions. Furthermore, the U-Net model achieved an AUC of 0.88 on the test set and an AUC of 0.81 on the validation set. This study effectively enhances the accuracy and reliability of landslide hazard identification by synergizing three modules—InSAR time-series analysis, machine learning classification, and deep learning segmentation—thereby providing a novel technical pathway for landslide disaster prevention in mountainous regions.