Work

Hybrid semantic segmentation framework for reinforced concrete bridge point clouds

Year: 2026

Type: article

Abstract: Accurate point cloud segmentation is critical for the geometric BIM reconstruction of Reinforced Concrete (RC) bridges. However, existing approaches still face challenges in real-world applications, such as training data scarcity and poor generalisability. To address these challenges, this paper proposes a hybrid semantic segmentation framework, HyBridgeSeg, combining the strengths of both learning-based segmentation models and domain-knowledge-based segmentation rules. HyBridgeSeg decomposes the multi-class segmentation task into several tractable subtasks, each carefully designed with a segmentation pipeline. Validated on mixed synthetic-real and fully real-world datasets, HyBridgeSeg consistently outperformed its learning-based baselines. Hyb-PTv3 achieved the best performance on the mixed dataset, with 98.49% OA, 98.43% mPr, and 96.37% mIoU, while improving mIoU from 57.26% to 76.54% in synthetic-to-real validation and from 67.84% to 87.06% in real-to-real validation. These results confirm the accuracy, transferability, and potential of the hybrid strategy for bridge point cloud semantic segmentation.

Source: Automation in Construction

DOI: 10.1016/j.autcon.2026.107255

Authors: Tao Yang, Yang Zou, Enrique del Rey Castillo, Ruidong Chang, Jian Zhong

Language: English

Topic: Infrastructure Maintenance and Monitoring

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDGs: Industry, innovation and infrastructure, Sustainable cities and communities

Open Access status: hybrid

Estimated APC paid: $5,490

Funders: University of Auckland, China Scholarship Council