高速公路结构裂缝三维形态重建方法OA
3D morphological reconstruction method for highway structural cracks
为解决高速公路结构裂缝三维重建中空间信息利用率低、形态表达不充分的问题,提出一种融合三维分割网络与等值面提取的结构裂缝三维形态重建方法.该方法利用三维探地雷达采集多通道数据,经预处理后基于3D窗口提取全剖面图谱特征,构建结构裂缝三维体数据集,在3D U-Net框架基础上,结合三维残差块与三维注意力机制,构建3 D SCMS-Net分割网络,以增强对裂缝空间连续特征和边界细节的表达能力.采用等值面提取算法对分割结果进行三维表面重建,实现裂缝三维形态表达.研究结果表明:在当前实验设置下,与二维切片输入对比模型相比,本文方法取得了更高的交并比(IoU)和更低的平均表面距离(ASD),表明基于三维体数据的处理方式有助于提升裂缝空间信息利用效率与形态分割效果;在不同天线频率及不同高速公路路段场景下,IoU稳定在0.75以上,ASD低于7.00,各指标变异系数均小于3%.现场钻芯验证显示,重建结果与芯样中裂缝走向及弯曲趋势基本一致.研究结论为高速公路结构裂缝的精准检测与科学养护决策提供技术支撑.
To address the low utilization of spatial information and insufficient morphological representation in the 3D reconstruction of highway structural cracks,this study proposes a 3D morphological reconstruction method integrating a 3D segmentation network and isosurface extraction.First,multi-channel data are collected using a 3D ground-penetrating radar(GPR)system.After preprocessing,full-profile spectral features are extracted using a 3D window to construct a 3D volumetric dataset of structural cracks.Then,based on the 3D U-Net framework,a task-oriented 3D SCMS-Net is developed by incorporating 3D residual blocks and a 3D attention mechanism to enhance the representation of crack spatial continuity and boundary details.Finally,an isosurface extraction algorithm is employed to reconstruct the 3D morphology of cracks from the segmentation results.The research results show that,under the current experimental setting,the proposed method achieves higher Intersection over Union(IoU)and lower Average Symmetric Surface Distance(ASD)than the comparison models using 2D slice-based input,indicating that 3D volumetric processing is beneficial for improving spatial information utilization and morphological segmentation of cracks.Under different antenna frequencies and highway section conditions,the IoU remains above 0.75,the ASD is below 7.00,and the coefficient of variation of each metric is less than 3%.Field core-drilling validation further shows that the reconstructed results exhibit high consistency with the crack direction and bending trend observed in the core samples.This research conclusion provides reliable technical support for the accurate detection of highway structural cracks and scientific maintenance decision-making.
张伟光;郑营运;张一鸣;申熙远;孟凡文
东南大学 交通学院,江苏 南京 211189东南大学 交通学院,江苏 南京 211189||中交天津港湾工程研究院有限公司,天津 300222||中交第一航务工程局有限公司,天津 300461东南大学 交通学院,江苏 南京 211189东南大学 交通学院,江苏 南京 211189中交天津港湾工程研究院有限公司,天津 300222||中交第一航务工程局有限公司,天津 300461
交通工程
三维探地雷达结构裂缝三维体数据集三维分割网络注意力机制等值面提取算法
3D ground-penetrating radarstructural cracks3D volumetric dataset3D segmentation networkattention mechanismisosurface extraction algorithm
《辽宁工程技术大学学报(自然科学版)》 2026 (5)
607-618,12
国家自然科学基金面上项目(52278443)云南省重大科技专项计划(202502AD080007)
评论