Page 61 - 2025年第56卷第9期

.
［ 15］ 李锋 . 基于无人机及深度学习的桥梁结构裂缝智能识别［D］ 长沙：湖南大学，2021.
.
［ 16］ 赵男 . 基于无人机图像的钢结构表面锈蚀和疲劳裂纹识别定位［D］ 天津：天津大学，2021.
［ 17］ CHAIYASARN K，BUATIK A，MOHAMAD H，et al. Integrated pixel-level CNN-FCN crack detection via photo⁃
grammetric 3D texture mapping of concrete structures［J］ Automation in Construction，2022，140：104388.
.
［ 18］ CHOW J K， LIU K， TAN P S， et al. Automated defect inspection of concrete structures［J］ Automation in Con⁃
.
struction，2021，132：103959.
［ 19］ YUAN C，XIONG B，LI X，et al. A novel intelligent inspection robot with deep stereo vision for three-dimensional
concrete damage detection and quantification［J］ Structural Health Monitoring，2022，21（3）：788-802.
.
/
［ 20］ HOU Q，ZHOU D，FENG J. Coordinate attention for efficient mobile network design［C］/2021 IEEE/CVF Confer⁃
ence of Computer Vision and Pattern Recognition （CVPR） . 2021.
［ 21］ ALOM M Z， YAKOPCIC C， TAHA T M， et al. Nuclei segmentation with recurrent residual convolutional neural
networks based U-Net （R2U-Net）［C］/IEEE National Aerospace and Electronics Conference. 2018.
/
［ 22］ BADRINARAYANAN V，KENDALL A，CIPOLLA R. Segnet：A deep convolutional encoder-decoder architecture
for image segmentation［J］ IEEE Transactions on Pattern Analysis and Machine Intelligence， 2017， 39（12）：
.
2481-2495.
［ 23］ ZOU Q，ZHANG Z，LI Q，et al. Deepcrack：Learning hierarchical convolutional features for crack detection［J］
.
IEEE Transactions on Image Processing，2018，28（3）：1498-1512.
［ 24］ WANG B， WANG S， QIU S， et al. CSU-Net： A context spatial U-Net for accurate blood vessel segmentation in
.
fundus images［J］ IEEE Journal of Biomedical and Health Informatics，2020，25（4）：1128-1138.

Three-dimensional reconstruction and defect quantitative identification method of hydraulic
concrete structure based on UAV images and deep learning

1，2，3 2 3 1 2 4

LI Yangtao ，ZHAO Haitao ，GU Chongshi ，WEI Yang ，BAO Tengfei ，XIANG Zhenyang

（1. School of Civil Engineering，Nanjing Forestry University，Nanjing 210037，China；2. College of Civil Engineering and

Transportation，Hohai University，Nanjing 210098，China；3. National Key Laboratory of Water Disaster Prevention，

Hohai University，Nanjing 210098，China；4. Sichuan Rural Water Conservancy Center，Chengdu 610072，China）

Abstract：During the service life of concrete structures，they are prone to defects such as cracks，spalling，and cal⁃

cium precipitation due to environmental erosion and multi-load coupling. Traditional manual inspection methods
suffer from drawbacks，including low detection efficiency，high potential risks，and a high rate of misjudgment，

making them difficult to implement in high-rise structural areas. To address these challenges，this paper proposes a

method for 3D reconstruction and defect quantification of concrete structures using multi-view stereo vision and deep
t
learning. Building on the U-Net3+ model，he paper introduces the coordinate attention mechanism and a classifica⁃
tion guidance module to enhance the model′s ability to focus on minute defect areas and improve the accuracy of

defect segmentation in complex，background-interfered environments. Recognizing that traditional machine vision
detection based on 2D images struggles with defect localization and accurate size quantification，a 3D reconstruction

method for concrete structures from drone images is developed. A visual 3D reconstruction framework is established to
perform reverse engineering calculations and restore the imaging process. Defects identified in 2D images are inte⁃
grated with the 3D real-world scene model，and a method for defect localization and quantification in concrete struc⁃

tures using multi-view stereo vision is proposed. Concrete beams and a concrete high dam are used as engineering
case studies to validate the accuracy of the proposed 3D reconstruction for concrete structures and the effectiveness of
fine defect identification and localization. The results demonstrate that the method proposed in this paper enables pre⁃
cise quantitative identification and spatial localization of defects in concrete structures，even under the complex back⁃

ground interference of UAV-based aerial imagery. This approach significantly improves detection efficiency and pro⁃
vides valuable technical support for long-term defect tracking and monitoring.

Keywords：hydraulic concrete structure；defect identification；drone aerial photography；machine vision；artificial

intelligence
（责任编辑：王 婧）
— 1165 —