KSCE Journal of Civil Engineering, Год журнала: 2024, Номер unknown, С. 100095 - 100095
Опубликована: Окт. 1, 2024
Язык: Английский
Construction and Building Materials, Год журнала: 2024, Номер 416, С. 135036 - 135036
Опубликована: Фев. 1, 2024
The performance of deep learning-based computer vision systems for road infrastructure assessment is hindered by the scarcity real-world, high-volume public datasets. Current research predominantly focuses on crack detection and segmentation, without devising end-to-end capable effectively evaluating most affected roads assessing out-of-sample performance. To address these limitations, this study proposes a dataset with annotations 7099 images 13 types defects, not only based cracks, confrontation development learning models. These are used to train compare YOLOv5 sub-models pure efficiency, standard object metrics, select optimum architecture. A novel post-processing filtering mechanism then designed, which reduces false positive detections 20.5%. Additionally, pavement condition index (ASPDI) engineered models identify areas in need immediate maintenance. facilitate decision-making administrations, software application created, integrates ASPDI, geotagged images, detections. This tool has allowed detect two sections critical repair. refined architecture validated open datasets, achieving mean average precision scores 0.563 0.570 RDD2022 CPRI, respectively.
Язык: Английский
Процитировано
14
Construction and Building Materials, Год журнала: 2024, Номер 436, С. 136733 - 136733
Опубликована: Июнь 7, 2024
Язык: Английский
Процитировано
6
Nondestructive Testing And Evaluation, Год журнала: 2024, Номер unknown, С. 1 - 68
Опубликована: Ноя. 22, 2024
With the advancement and expansion of power system technology, ensuring safe stable operation transmission lines has become increasingly crucial. Traditional manual inspection methods often suffer from low efficiency high risks. In recent years, rapid development unmanned aerial vehicle (UAV) technology provided a new solution for inspection. UAV offers advantages such as flexibility, wide coverage, relatively overall costs, effectively improving efficiency, reducing mitigating personnel safety A comprehensive review current state inspections in systems is this study, introducing main sensing technologies applications inspection, including visible light cameras, infrared thermal imaging, depth compound eye radar, X-rays, ultrasonic sensors. It also discusses challenges faced by systems, complexity data processing, lack automation, absence regulatory frameworks. Furthermore, it put forward future trends, multi-sensor fusion, edge computing, cloud computing centres, multi-UAV collaboration, cooperation various clusters. This work aims to serve reference research application systems.
Язык: Английский
Процитировано
6
Automation in Construction, Год журнала: 2024, Номер 170, С. 105899 - 105899
Опубликована: Дек. 11, 2024
Язык: Английский
Процитировано
4
Journal of Real-Time Image Processing, Год журнала: 2025, Номер 22(2)
Опубликована: Фев. 19, 2025
Язык: Английский
Процитировано
0
Journal of Infrastructure Intelligence and Resilience, Год журнала: 2025, Номер unknown, С. 100144 - 100144
Опубликована: Фев. 1, 2025
Язык: Английский
Процитировано
0
Measurement, Год журнала: 2025, Номер unknown, С. 117344 - 117344
Опубликована: Март 1, 2025
Язык: Английский
Процитировано
0
Computers, materials & continua/Computers, materials & continua (Print), Год журнала: 2024, Номер 79(1), С. 1679 - 1703
Опубликована: Янв. 1, 2024
The detection of crack defects on the walls road tunnels is a crucial step in process ensuring travel safety and performing routine tunnel maintenance.The automatic accurate cracks surface key to improving maintenance efficiency tunnels.Machine vision technology combined with deep neural network model an effective means realize localization identification tunnels.We propose complete set inspection methods for identifying as solution problem difficulty during manual maintenance.First, equipment applied real-time acquisition high-definition images designed.Images are acquired based designed equipment, where containing manually identified selected.Subsequently, training validation sets used construct obtained images, whereas regions pixels finely labeled.After that, area sensing module proposed you only look once version 7 coordinate attention mechanism (CA-YOLO V7) locate images.Only subimages sent multiscale semantic segmentation extraction which belong DeepLab V3+ network.The precision recall region our method 82.4% 93.8%, respectively.Moreover, mean intersection over union (MIoU) pixel accuracy (PA) values achieving pixel-level 76.84% 78.29%, respectively.The experimental results dataset show that two-stage outperforms other state-of-the-art models detection.Based method, captured can at speed ten frames/second, reach 0.25 mm, meets requirements actual project.The CA-YOLO V7 enables precise belongs under different environmental lighting conditions tunnels.The improved lightweighting able extract morphology given more quickly while maintaining accuracy.The established combines defect first time, realizing complex environments, capable determining size physical system after camera calibration.The trained has high be extended embedded computing devices assessment repair damaged areas types tunnels.
Язык: Английский
Процитировано
2
Measurement, Год журнала: 2024, Номер unknown, С. 116486 - 116486
Опубликована: Дек. 1, 2024
Язык: Английский
Процитировано
2
KSCE Journal of Civil Engineering, Год журнала: 2024, Номер unknown, С. 100095 - 100095
Опубликована: Окт. 1, 2024
Язык: Английский
Процитировано
1