基于AHC方法的地下洞室岩体结构面自动识别方法

Automatic identification method for discontinuities in underground rock masses based on the AHC method

摘要

地下洞室岩体中的结构面对工程稳定性和施工安全至关重要。然而，在复杂地质条件下，现有识别与聚类方法在应对数据分布不均、噪声干扰等方面仍存在精度下降与关键结构面遗漏的问题。为此，采用虚拟多目相机采集岩体图像，结合SFM-MVS技术构建高精度三维点云模型，在提升数据采集效率的同时，规避了人工测量的安全隐患。提出自适应区域生长算法，有效实现复杂地质条件下独立结构面的精确识别。结合岩体产状数据服从Fisher分布的统计特性，设计了基于统计检验的自适应层次聚类算法(Adaptive Hierarchical Clustering, AHC),自动确定结构面最优分组数，克服了传统方法中预设聚类簇数或阈值的限制。在案例中，该方法成功处理包含63.9万点的点云数据，耗时37 s,展现出优异的处理效率。聚类结果表明，AHC算法识别精度为96.43%,分别较主流方法提升8.5%和14.3%。同时，自适应区域生长算法在尖角和裂隙交汇等复杂边界区域亦保持较高的识别完整性。

Abstract

Discontinuities in underground rock masses are critical for engineering stability and construction safety. However, under complex geological conditions, existing method are often hindered by limited recognition and clustering accuracy, particularly in the presence of uneven data distribution or noise. A virtual multi-camera system is utilized to acquire rock surface imagery, and high-precision 3 D point cloud models are generated using Structure-from-Motion and Multi-View Stereo(SFM-MVS) techniques. This approach effectively mitigates the safety risks associated with manual measurements and enhances data acquisition efficiency.To process the point cloud data, a self-adaptive region-growing algorithm is developed, enabling accurate identification of independent structural planes under intricate geological conditions. Additionally, an Adaptive Hierarchical Clustering(AHC) algorithm is proposed based on statistical analysis and the Fisher distribution, allowing the optimal number of structural plane groups to be determined automatically. This addresses the limitations of conventional method that rely on predefined cluster numbers or thresholds.The proposed framework was validated using a dataset comprising 639000 points, which was successfully processed in 37 seconds, demonstrating high computational efficiency. The AHC algorithm achieved a recognition accuracy of 96.43%, outperforming conventional method by 8.5% and 14.3%, respectively. Furthermore, the self-adaptive region-growing algorithm exhibited strong completeness in identifying complex boundary regions such as sharp edges and fissure intersections.

关键词

地下洞室 / 岩体不连续结构面 / 三维点云 / 产状最优分组 / 层次迭代聚类算法

Key words

underground cavern / rock mass discontinuities / 3 D point cloud / optimal attitude grouping / hierarchical iterative clustering algorithm

Author summay

王志敏(1992—), 男, 工程师, 学士, 主要从事水利水电智能建造技术研究。E-mail: [e-mail]

基金资助

上海勘测设计研究院有限公司科研项目(2023XX(83)-001)