同济大学学报（自然科学版）2026，Vol.54Issue(8)：1155-1166,12.DOI:10.11908/j.issn.0253-374x.25160
隧道围岩超欠挖自注意力机制的高斯混合模型点云配准方法
Self-Attention-Driven Gaussian Mixture Model Point-Cloud Registration Method for Evaluating Over-and Under-Excavation in Tunnel Surrounding Rock
摘要
Abstract
The over-excavation and under-excavation values of surrounding rock are important indicators for tunnel construction quality and safety control.Traditional total station measurements of tunnel surrounding rock are inefficient for detecting over-and under-excavation.Three-dimensional(3D)laser scanning technology,however,enables efficient acquisition of spatial point cloud data.Nevertheless,conventional tunnel point cloud ellipse fitting methods are difficult to apply to the extraction of over-and under-excavation values in multi-center circular tunnel surrounding profiles.To address this issue,this paper proposes a point cloud registration method for tunnel surrounding rock over-and under-excavation based on a Gaussian mixture model(GMM)with a self-attention mechanism.The method first constructs a 3D model based on the tunnel design profile and transforms it into a design point cloud.Next,for the measured point cloud of the tunnel surrounding rock profile,voxel filtering and a density-based spatial clustering algorithm are applied for down-sampling and denoising.The adaptability of the method to tunnel surrounding rock point cloud profiles is verified,and optimal parameters are obtained.Finally,a Gaussian mixture model based on a self-attention mechanism is proposed.After registration of the design and measured point clouds,the method effectively extracts the over-and under-excavation values of the tunnel surrounding rock.The proposed method was successfully applied in the construction of the Tianshengqiao Tunnel on the Lincang-Shuangjiang Expressway in Yunnan.The maximum error in over-and under-excavation extraction was 2.01 mm,and the volume error percentage was 1.97％,meeting the precision requirements for engineering applications.This approach provides effective technical support for the rapid extraction of full cross-sectional over-and under-excavation values in tunnel surrounding rock.关键词
隧道工程/围岩/超欠挖/点云生成/自注意力机制/高斯混合模型/点云配准Key words
tunnel engineering/surrounding rock/overbreak and underbreak/point cloud generation/self-attention mechanism/Gaussian mixture model(GMM)/point cloud registration分类
交通工程引用本文复制引用
韩钊,谢雄耀,唐亘跻,李培锋..隧道围岩超欠挖自注意力机制的高斯混合模型点云配准方法[J].同济大学学报（自然科学版）,2026,54(8):1155-1166,12.基金项目
国家重点研发计划资助(2023YFC3806705) （2023YFC3806705）
国家自然基金重点项目资助(52038008、52378408) （52038008、52378408）
云南交投科技创新计划项目(YCIC-YF-022-01) （YCIC-YF-022-01）
