基于最小割优化算法的桥梁群桩声呐点云提取方法

Minimum Cut Optimization Algorithm for Sonar Point Cloud Data Segmentation and Extraction of Bridge Pile Groups

桥梁水下基础对于涉水桥梁结构的安全至关重要,获取水下基础完整形态是非接触数字检测结构表观损伤和判断安全状况的重要前提。与桥梁上部结构不同,当前获取水下结构形态的手段单一,基于水下三维数据的桥梁下部结构损伤识别评估研究仍为起步阶段,多为针对水下图像的二维识别评估和面向水下三维结构数字模型的定性评判。为了解决这一问题,提出了一种面向桥梁水下三维声呐点云的水下群桩空间形态提取方法。该方法基于点云最小割算法,针对水下复杂环境群桩声呐点云稀疏性与结构相似性特征,引入迁移学习思想,定义源桩基与目标桩基,并提出源桩基到目标桩基参数迁移方法,迁移参数包括先验点与共享参数;进而,提出基于圆环边界拟合方法,最终实现对复杂水下环境中桥梁桩基点云的完整提取。该方法的可行性与适用性在芜湖长江大桥水下桩基提取中得到验证。结果表明:桩基提取精度可达0.725,召回率和F_(1)值分别为0.974和0.835,表明所提方法可有效稳定地分割提取桩基,并较好保留了桩基边缘的完整性,可为后续桩基表观损伤的识别及量化提供精准数据支持。

Underwater bridge foundations are critical to the safety of bridge structures.Obtaining an intact morphology of the underwater foundation is an essential prerequisite for the non-contact digital detection of apparent damage to a bridge structure and subsequent determination of its safety.In contrast to bridge superstructures,there are currently few methods for obtaining the morphology of underwater substructures,and research on the damage identification and assessment of bridge substructures based on underwater three-dimensional data is in its infancy.Most methods conduct a two-dimensional assessment of underwater images and employ qualitative judgments to create three-dimensional structural digital models.To overcome such reliance on qualitative judgement,a spatial morphology extraction approach for underwater pile groups based on underwater three-dimensional sonar point cloud data(USPCD)is presented in this article.Employing the point-cloud minimum cut algorithm,the proposed method introduces the idea of migration learning,defines the source and target pile bases,and utilizes a parameter migration method from the source to the target pile bases to capture the sparsity and structural similarity of the pile group sonar point cloud.The migration parameters include a priori points and shared parameters,and a circular boundary fitting method is applied to realize the complete extraction of the pile group point cloud from the complex underwater environment.The feasibility and applicability of the proposed method were verified by extracting the underwater pile foundation of the Wuhu Yangtze River Bridge.The results show that the accuracy of pile base extraction can reach 0.725,with recall and F_(1) scores of 0.974 and 0.835,respectively.This indicates that the proposed method can effectively and stably segment and extract pile groups while retaining the integrity of the group edges,which can provide accurate data support for the subsequent identification and quantification of any apparent damage.