融合无人机InSAR技术的滑坡体变形PointNet++识别方法

甘肃林业职业技术大学 测绘工程学院,甘肃 天水 741020

无人机InSAR技术;滑坡体;PointNet++;变形识别;三维点云模型;点云变形场

A PointNet++Based Identification Method for Landslide Deformation by Integrating UAV-InSAR Technology
SHI Qian,RUAN Guojie

Gansu Forestry Vocational and Technical University, School of Surveying and Mapping Engineering, Tianshui 741020, China

UAV InSAR technology; Landslide mass; PointNet++; Deformation identification; 3D point cloud model; Point cloud deformation field

DOI: 10.13512/j.hndz.2026.04.17

备注

现有滑坡体变形识别方法多侧重于时序预测或低维分析,难以有效捕捉和利用滑坡体在三维空间中的复杂变形结构与局部细节特征,导致识别精度与可靠性受限,为此研究融合无人机InSAR技术的滑坡体变形PointNet++识别方法。基于无人机InSAR技术,反演得到滑坡体变形量,经地理编码获取滑坡体变形场,及时掌握滑坡体全面且详细的变形信息;通过机载激光雷达点云数据,建立滑坡体三维点云模型;通过融合滑坡体变形场与三维点云模型,生成无缝滑坡体点云变形场,作为PointNet++的输入,通过提取点云数据的局部细节与全局结构特征,输出滑坡体变形识别结果。实验证明:该方法可有效获取滑坡体变形场;该方法可有效建立滑坡体三维点云模型,生成无缝点云变形场,且本文方法的总体识别准确率高达95%,精准完成滑坡体变形识别。
The existing methods for identifying landslide deformation mainly focus on temporal prediction or low-dimensional analysis, which makes it difficult to effectively capture and utilize the complex deformation structure and local details features of landslide masses in three-dimensional space, resulting in limited recognition accuracy and reliability. Therefore, this research proposes a PointNet++identification method for landslide deformation that integrates UAV-InSAR technology. Based on unmanned aerial vehicle InSAR technology, the deformation amount of the landslide masses is inverted, and the deformation field is obtained geocoding to timely grasp the comprehensive and detailed deformation information of the landslide body; Establish a three-dimensional point cloud model of the landslide body using airborne LiDAR point cloud data; By integrating the deformation field of the landslide masses with the three-dimensional point cloud model, a seamless landslide mass point cloud deformation field is generated as input to PointNet++. By extracting local details and global structural features of the point cloud data, the landslide masses deformation recognition results are output. Experimental results verify that this method can effectively obtain the deformation field of landslide masses. This method can effectively establish a three-dimensional point cloud model of landslide masses, generate seamless point cloud deformation field, and the overall recognition accuracy of this method is as high as 95%, accurately completing landslide masses deformation recognition.
·