Abstract:Advances in Structural Engineering, Ahead of Print. Due to the influence of various factors, such as complex environments and sustained load effects, the long-term service life of bridge structures will lead to a gradual deterioration in performance. Therefore, bridge health monitoring is of utmost importance, and component identification is a crucial step in evaluating the overall structural integrity of bridges. With the advancement of deep learning algorithms, semantic segmentation methods can effectively classify and identify bridge components in complex environments, thereby facilitating the assessment of their state. Nevertheless, the conventional methods for segmenting bridge components suffer from drawbacks such as intensive computation, inadequate feature extraction, and low segmentation accuracy, failing to meet the requirements of current bridge health monitoring. Consequently, this paper proposes a bridge component segmentation method based on an improved DeepLabV3 + model, named the DeepLabV3-MS, which is based on an enhanced DeepLabV3 + model. This method utilizes MobileNetV2 as the backbone network to reduce parameter count and improve the computational speed of the model. The Strip Pooling (SP) is also integrated into ASPP, known as SP_ASPP, to enhance the capture of more comprehensive contextual information. Additionally, the Multi-scale Channel Attention Mechanism (MS-CAM) is incorporated to enhance the integration efficiency of multi-semantic and multi-scale features. The results indicate that compared with the original DeepLabV3 + model, the Mean Intersection over Union and Mean Pixel Accuracy of the DeeplabV3-MS model increased by 5.90%, and 4.92%, respectively. Furthermore, in comparison to the classic models PSPNet and U-Net, DeeplabV3-MS demonstrated an increase of 19.50% and 8.88% in MIoU and MPA, respectively, as well as 13.50% and 5.34%, respectively. The proposed method has demonstrated superior performance across various evaluation metrics, exerting a significant impact on the health monitoring and safety assessment of bridge components. Furthermore, it offers valuable technical support for research and applications in related fields.

Development of parametric bridge BIM and PCD generation algorithms and PCD-based member segmentation

Automatic Geometric Digital Twin of Box Girder Bridge Using a Laser-Scanned Point Cloud

Automated Reconstruction of Parametric BIM for Bridge Based on Terrestrial Laser Scanning Data

Automated geometric digital twinning of bridges from segmented point clouds by parametric prototype models

Automated semantic segmentation of bridge point cloud based on local descriptor and machine learning

Analysis of Railway Bridge Girder Installation and Construction Technology--Based on BIM Technology

A bridge point cloud databank for digital bridge understanding

Data-Driven Modeling Algorithms for Cable-Stayed Bridges Considering Mechanical Behavior

A parametric BIM approach to foster bridge project design and analysis

A Robust Fast Bridging Algorithm for Laser Cutting

Semi-automated Generation of Geometric Digital Twin for Bridge Based on Terrestrial Laser Scanning Data

Automatic detection method for verticality of bridge pier based on BIM and point cloud

Automated bridge component recognition from point clouds using deep learning

A Deep-Learning-Based Bridge Damaged Object Automatic Detection Model Using a Bridge Member Model Combination Framework

Semi-automated generation of parametric BIM for steel structures based on terrestrial laser scanning data

Robust localization of shear connectors in accelerated bridge construction with neural radiance field

Parametric Design and Application of Assembled Steel Plate Composite Beam Bridge Based on BIM Technology

Parametric modeling and engineering application of the high-speed continuous beam bridge based on Building Information Modeling

A parametric approach for creating finite element models from point clouds for steel superstructure components in steel girder bridges

Bridge component segmentation for health monitoring an enhanced DeepLabV3+ model with lightweight network and multi-scale channel attention mechanism