Cited by Weakly‐supervised structural component segmentation via scribble annotations

Self‐training with Bayesian neural networks and spatial priors for unsupervised domain adaptation in crack segmentation DOI

Pang-jo CHUN,

T. Kikuta

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2024, Volume and Issue: 39(17), P. 2642 - 2661

Published: July 29, 2024

Abstract This study proposes a novel self‐training framework for unsupervised domain adaptation in the segmentation of concrete wall cracks using accumulated crack data. The proposed method incorporates Bayesian neural networks uncertainty estimation pseudo‐labels, and spatial priors screening noisy labels. Experiments demonstrate that approach achieves significant improvements F1 score. Comparing scores, DeepLabv3+ U‐Net showed performance 0.0588 0.1501, respectively, after adaptation. Furthermore, integration Stable Diffusion few‐shot image generation enhances by 0.0332. enables high‐precision with as few 100 target images, which can be easily obtained at site, reducing cost model deployment infrastructure maintenance. also investigates optimal number iterations based on score, providing insights practical implementation. contributes to development efficient automated structural health monitoring AI.

Language: Английский

Citations

Deep learning-based corrosion inspection of long-span bridges with BIM integration DOI

K. Hattori,

Keiichi Oki,

Aya Sugita

et al.

Heliyon, Journal Year: 2024, Volume and Issue: 10(15), P. e35308 - e35308

Published: July 30, 2024

Infrastructure operation and maintenance is essential for societal safety, particularly in Japan where the aging of infrastructures built during period high economic growth advancing. However, there are issues such as a shortage engineers inefficiencies work, requiring improvements efficiency automation their resolution. Nevertheless, still many current procedures bridge inspections. Usually, inspection check damage on bridges through close visual inspections at site, then photograph damaged parts, measure size by touch, create report. A three-dimensional representation, considering front back structural elements, needed identifying damage, necessitating creation multi-directional drawings. this process labor-intensive prone to errors. Furthermore, due lack uniformity records, it challenging refer past histories. Especially long bridges, without resolving issues, required labor number mistakes could exceed acceptable limits, making proper management difficult. Therefore, study, we developed method automatically measuring position area corroded parts capturing images lower surface stiffening girder using vehicle utilizing image diagnosis technology. By integrating these results into 3D model called BIM (Building Information Modeling), becomes possible manage more efficiently. We verified actual confirmed its effectiveness.

Language: Английский

Citations

Study on the automated characterization of particle size and shape of stacked gravelly soils via deep learning DOI

Jian Gong, Ziyang Liu,

Jiayan Nie

et al.

Acta Geotechnica, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 4, 2025

Language: Английский

Citations

A structure‐oriented loss function for automated semantic segmentation of bridge point clouds DOI

Chao Lin,

Shuhei Abe,

Shitao Zheng

et al.

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 12, 2025

Abstract Focusing on learning‐based semantic segmentation (SS) methods for bridge point cloud data (PCD), this study proposes a structure‐oriented concept (SOC) with training focused the spatial distribution patterns of components, including both horizontally absolute location each component and its vertically relative position compared other components. Then loss (SOL) function, which embodies core SOC, is defined accordingly, it to five cutting‐edge functions collected PCD dataset. In contrast limitations functions, SOL significantly improves overall evaluation metrics accuracy (6.53%) mean intersection over union (mean IoU: 8.67%). The IoU category “others” improved by 8.44%, very important automating time‐consuming denoising process. Furthermore, demonstrated robustness SOC reveal great potential improve performance SS models.

Language: Английский

Citations

Implementation of explanatory texts output for bridge damage in a bridge inspection web system DOI

Pang-jo CHUN,

Hong-Hu Chu,

Kota Shitara

et al.

Advances in Engineering Software, Journal Year: 2024, Volume and Issue: 195, P. 103706 - 103706

Published: June 21, 2024

Language: Английский

Citations

A machine vision‐based intelligent segmentation method for dam underwater cracks using swarm optimization algorithm and deep learning DOI

Yantao Zhu, Xinqiang Niu,

Jinzhang Tian

et al.

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 2, 2024

Abstract Ensuring the safety of water networks is a research hotspot in current conservancy industry, and dams are an important part. However, over time, dam prone to varying degrees aging disease, most which structural cracks. If they cannot be discovered repaired normal operation will affected, even catastrophic accidents such as failure occur. complex backgrounds blurred images can easily lead misjudgments by machine vision detection models, high‐efficiency accurate evaluation technology urgently needed. This paper combines deep semantic segmentation network model hyperparameters optimization algorithm propose data‐intelligent perception method underwater cracks driven knowledge coupling. Taking concrete face rockfill example, effectiveness verified using vehicle carrier. Experimental results indicate that developed achieves intersection‐union ratio 0.9301, precision rate 0.9678, 0.9472, recall 0.9577 test set. shows constructed has high crack fine performance. In addition, better performance different scenes, further illustrates method.

Language: Английский

Citations

A graph attention reasoning model for prefabricated component detection DOI

Manxu Zhou,

Guanting Ye,

Ka‐Veng Yuen

et al.

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 2, 2025

Abstract Accurately checking the position and presence of internal components before casting prefabricated elements is critical to ensuring product quality. However, traditional manual visual inspection often inefficient inaccurate. While deep learning has been widely applied quality components, most studies focus on surface defects cracks, with less emphasis structural complexities these components. Prefabricated composite panels, due their complex structure—including small embedded parts large‐scale reinforcing rib—require high‐precision, multiscale feature recognition. This study developed an instance segmentation model: a graph attention reasoning model (GARM) for component detection, concrete panels. First, dataset was constructed address shortage existing data provide sufficient samples training network. Subsequently, after self‐built ablation experiments comparative tests were conducted. The GARM demonstrated superior performance in terms detection speed lightweighting. Its accuracy surpassed other models, mean average precision (mAP 50 ) 88.7%. confirms efficacy reliability detecting

Language: Английский

Citations

Seismic Damage Quantification of RC Short Columns from Crack Images Using the Enhanced U-Net DOI

Zefan Chen,

Qian Chen, Zhangjun Dai

et al.

Buildings, Journal Year: 2025, Volume and Issue: 15(3), P. 322 - 322

Published: Jan. 22, 2025

It is of great importance to quantify the seismic damage reinforced concrete (RC) short columns since they often experience severe due likely excessive shear deformation. In this paper, quantification method RC under earthquakes proposed based on crack images and enhanced U-Net. To end, short-column specimens were prepared tested cyclic loading. The force-displacement hysteresis curves obtained quantitatively calculate indicator column energy criterion. At same time, surfaces taken by smartphones using partition photographing scheme image stitching algorithm. widely used U-Net was adding a double attention mechanism segment cracks in images. results demonstrate that it has better accuracy terms recognizing tiny compared original By analysis, information further extracted from investigate development columns. Finally, correlations between criterion loading analyzed, showing highest correlation exists total area. normalized area, i.e., ratio area corresponding monitored surface, defined when utilizing for assessment.

Language: Английский

Citations

Transformer‐based framework for accurate segmentation of high‐resolution images in structural health monitoring DOI

Maryam Azimi, T.Y. Yang

Computer-Aided Civil and Infrastructure Engineering, Journal Year: 2024, Volume and Issue: unknown

Published: April 21, 2024

Abstract High‐resolution image segmentation is essential in structural health monitoring (SHM), enabling accurate detection and quantification of components damages. However, conventional convolutional neural network‐based methods face limitations real‐world deployment, particularly when handling high‐resolution images producing low‐resolution outputs. This study introduces a novel framework named Refined‐Segment Anything Model (R‐SAM) to overcome such challenges. R‐SAM leverages the state‐of‐the‐art zero‐shot SAM generate unlabeled masks, subsequently employing DEtection Transformer model label instances. The key feature contribution its refinement module, which improves accuracy masks generated by without need for extensive data annotations fine‐tuning. effectiveness proposed was assessed through qualitative quantitative analyses across diverse case studies, including multiclass segmentation, simultaneous tracking, 3D reconstruction. results demonstrate that outperforms convolution models with mean intersection‐over‐union 97% boundary 87%. In addition, achieving high coefficients determination target‐free tracking studies highlights versatility addressing various challenges SHM.

Language: Английский

Citations

DUCTNet: An Effective Road Crack Segmentation Method in UAV Remote Sensing Images Under Complex Scenes DOI

Lixiang Sun, Yixin Yang,

Zaichun Yang

et al.

IEEE Transactions on Intelligent Transportation Systems, Journal Year: 2024, Volume and Issue: 25(9), P. 12682 - 12695

Published: Sept. 1, 2024

Road crack detection in complex scenarios is challenged by vehicles, traffic facilities, road printed signs and fine cracks. In order to better solve these problems, a novel dense nested depth U-shaped structure for image segmentation network named DUCTNet proposed. Firstly, designed combining the superior performance of Unet $++$ deep U2Net. This improves ability model extract features depth. Second, competitive fusion feature extraction block It dissimilarity between cracks background fusion. Then, high-density attention mechanism method enhances contextual sensitive information both horizontally vertically increasing density. Finally, achieves best results comparison tests with eight state-of-the-art specialized networks self-built datasets four public datasets. addition, excellent real tests, which proves that can provide engineers technicians means detecting

Language: Английский

Citations