Abstract

Rockfall has become one of the deadliest geohazards in Southwest China and how to comprehensively and effectively assess rockfall hazards is an urgent challenge to overcome. Additionally, comprehensive characterization of fractures on rock mass outcrops is a prerequisite for detecting potential rockfall. In this paper, an image and point cloud-based data fusion technique is applied to characterize regional rock mass fractures. Firstly, the performances of three classical computer vision algorithms are compared and SegFormer is selected as the appropriate base model for fracture detection. After that, according to the coordinate projection transformation criterion, the detected fractures are mapped to the point cloud. The parameter information obtained through fracture characterization is used to develop a representative three-dimensional discrete fracture network (3D-DFN) and then according to the results of the volume distribution of rock blocks, the three frequencies (high-frequency, medium-frequency, and low-frequency) of rockfall events are numerically simulated to obtain the characteristic information of rockfall trajectories. Finally, based on the characteristic information of rockfall trajectories and the GIS platform, the risk of rockfall hazards with three frequencies is evaluated and analyzed. This paper provides a new way for geologists to assess the risk of rockfall hazards and propose reasonable rockfall hazard prevention schemes.

中文翻译：

---

基于数据融合技术的裂缝表征及其在落石灾害评估中的应用

摘要

落石已成为西南地区最严重的地质灾害之一，如何全面、有效地评估落石灾害是亟待克服的挑战。此外，岩体露头裂缝的综合表征是检测潜在落石的先决条件。本文采用基于图像和点云的数据融合技术来表征区域岩体裂缝。首先，比较了三种经典计算机视觉算法的性能，并选择SegFormer作为裂缝检测的合适基础模型。之后，根据坐标投影变换准则，将检测到的裂缝映射到点云上。利用裂缝表征获得的参数信息建立具有代表性的三维离散裂缝网络（3D-DFN），然后根据岩块体积分布结果，将三个频率（高频、中频、对落石事件进行数值模拟，获取落石轨迹特征信息。最后，基于落石轨迹特征信息和GIS平台，对3个频率的落石灾害风险进行了评估和分析。本文为地质学家评估落石灾害风险并提出合理的落石灾害防治方案提供了新的途径。