地面LiDAR探测滑坡体小形变场的关键技术研究

The small surface deformation is one precursor signal of landslide, and the detection of three-dimensional deformation field with high resolution plays a significant role in disaster monitoring and prewarning. Ground-based LiDAR has become a new means for monitoring the small deformation due to its ability to obtain three-dimensional point clouds with high accuracy and density. However, there are still several problems that need to be solved urgently in the processing of point clouds from ground-based LiDAR in a landslide scene, such as big difficulty, low accuracy and low degree of automation. The project will be carried out following the outline of “Registration – Ground Points Extraction – Deformation Detection”. Three novel methods are considered to address the technical challenges, including the assessment of the reliability of each registration primitive, the robust extraction of ground points, and the computation of nonlinear deformations with a high accuracy. Firstly, a dynamic model fusing multiple weights is proposed to better estimate the reliability of each registration primitive, and to match point clouds with different geo-references. Then, a Conditional Random Field based on a Transfer Random Forest is developed to robustly extract ground points. Finally, the project proposes an elastic deformation energy model with constraints of terrain features， and to calculating three-dimensional non-linear small deformation by the finite element method. By these methods, the project aims to establish a methodology for precisely processing of point clouds, and to robustly detect the small deformation in a landslide scene. The outcome of this project is supposed to improve the ability of ground-based LiDAR to monitor the landslide, and to provide a good foundation for the study of disaster mechanism and disaster prevention and mitigation.

地表小形变是滑坡灾害前兆信号之一，精细探测其三维小形变场对灾害监测预警具有重要意义。地面LiDAR凭借其三维点云的高精度、高密度特性，已成为滑坡体小形变场精细监测的新手段。目前，滑坡场景地面LiDAR点云处理还存在难度大、精度差、自动化程度低等不足，是当前亟待解决的问题。本项目以“点云配准—>地面点云提取—>形变场探测”为研究主线，针对配准基元可靠性评估质量低、地面点云稳健提取难、非线性形变计算精度差等技术瓶颈，研究多类型权重动态组合模型，高质量评估基元可靠性，实现点云高精度配准；发展迁移随机森林的条件随机场模型，稳健提取地面点云；提出地形特征约束的弹性形变能量模型，并采用有限元方法高精度解算三维非线性小形变场，从而建立一套滑坡场景地面LiDAR点云高精度处理方法，实现滑坡小形变场稳健探测。本项目研究成果有望提高地面LiDAR监测滑坡灾害的能力，为灾害机理研究及防灾减灾提供重要的技术支撑。

地表小形变是滑坡灾害前兆信号之一，如何准确探测滑坡体小形变场，对灾害监测预警具有重要意义。地面LiDAR通过获取滑坡区域多时相高精度、高密度三维点云，能够实现厘米级甚至更小的形变场探测，受到国内外学者的广泛关注。其中，地面LiDAR点云高质量获取、多时相点云配准、地面点云提取和高精度形变场计算是从点云中探测小形变场的重要关键，也是地面LiDAR滑坡监测中面临的重要问题。因此，针对以上问题，本项目开展了系列研究，其中重要研究进展如下：1）研究了一种顾及点对可靠性的多时相地面LiDAR点云高精度配准方法，通过构建点对可靠性多类型因子动态评估组合模型，定量确定点对的可靠性程度，有效提高了多时相点云精配准质量；2）发展了一种基于机器学习的滑坡体多时相激光点云滤波分类新方法，通过挖掘缓慢滑坡体多时相地形之间的强相关性，极大地降低了对分类器泛化能力的要求；3）发展了一种自适应局部区域C2M高精度形变场计算方法，有效解决了滑坡场景复杂地形形态和地面LiDAR点云漏洞对形变计算的影响；4）提出了基于多目标优化的滑坡场景地面LiDAR观测网优化方法，进行滑坡地面LiDAR观测网科学布设，以提高地面LiDAR点云采集质量。研究成果突破了地面LiDAR观测网优化、多时相点云配准、多时相点云滤波分类和小形变场计算等关键技术，为实现复杂滑坡场景地面LiDAR形变场高质量探测提供了技术支撑，并在三峡库区大树镇滑坡和川藏铁路洛隆滑坡进行了实验。按照既定的研究计划，项目圆满完成了规定的研究内容。在项目执行中，项目组在Landslides、ISPRS等期刊上发表SCI标注论文4篇，其中3篇中科院1区Top、1篇中科院2区Top，获批软件著作权1项，申请国家发明专利1项，参加国内外学术会议6次，获测绘科技进步奖二等奖1次。