Citation: | Wang P F, Li Y T, Huang Y Y, et al. Defects detection for cable surface of cable-stayed bridge based on improved YOLOv5s network[J]. Opto-Electron Eng, 2024, 51(5): 240028. doi: 10.12086/oee.2024.240028 |
[1] | 颜东煌, 郭鑫. 斜拉索损伤对在役斜拉桥体系可靠度的影响[J]. 中南大学学报(自然科学版), 2020, 51(1): 213−220. doi: 10.11817/j.issn.1672-7207.2020.01.024 Yan D H, Guo X. Influence of damage of stay cables on system reliability of in-service cable-stayed bridges[J]. J Central South Univ (Sci Technol), 2020, 51(1): 213−220. doi: 10.11817/j.issn.1672-7207.2020.01.024 |
[2] | Xu F Y, Dai S Y, Jiang Q S, et al. Developing a climbing robot for repairing cables of cable-stayed bridges[J]. Autom Constr, 2021, 129: 103807. doi: 10.1016/J.AUTCON.2021.103807 |
[3] | 李少波, 杨静, 王铮, 等. 缺陷检测技术的发展与应用研究综述[J]. 自动化学报, 2020, 46(11): 2319−2336. doi: 10.16383/j.aas.c180538 Li S B, Yang J, Wang Z, et al. Review of development and application of defect detection technology[J]. Acta Autom Sin, 2020, 46(11): 2319−2336. doi: 10.16383/j.aas.c180538 |
[4] | Zheng Z L, Yuan X Q, Huang H H, et al. Mechanical design of a cable climbing robot for inspection on a cable-stayed bridge[C]//Proceedings of the 13th World Congress on Intelligent Control and Automation, 2018: 1680–1684. https://doi.org/10.1109/WCICA.2018.8630709. |
[5] | Ho H N, Kim K D, Park Y S, et al. An efficient image-based damage detection for cable surface in cable-stayed bridges[J]. NDT E Int, 2013, 58: 18−23. doi: 10.1016/j.ndteint.2013.04.006 |
[6] | Li X K, Gao C, Guo Y C, et al. Cable surface damage detection in cable-stayed bridges using optical techniques and image mosaicking[J]. Opt Laser Technol, 2019, 110: 36−43. doi: 10.1016/j.optlastec.2018.07.012 |
[7] | 赵鹤, 杨晓洪, 杨奇, 等. 融合注意力机制的金属缺陷图像分割方法[J]. 光电子·激光, 2021, 32(4): 403−408. doi: 10.16136/j.joel.2021.04.0411 Zhao H, Yang X H, Yang Q, et al. Metal defect image segmentation algorithm combined with attention mechanism[J]. J Optoelectron Laser, 2021, 32(4): 403−408. doi: 10.16136/j.joel.2021.04.0411 |
[8] | Pan G, Zheng Y X, Guo S, et al. Automatic sewer pipe defect semantic segmentation based on improved U-Net[J]. Autom Constr, 2020, 119: 103383. doi: 10.1016/j.autcon.2020.103383 |
[9] | Xiao Y Z, Tian Z Q, Yu J C, et al. A review of object detection based on deep learning[J]. Multimed Tools Appl, 2020, 79(33-34): 23729−23791. doi: 10.1007/s11042-020-08976-6 |
[10] | 赵宏伟, 郑嘉俊, 赵鑫欣, 等. 基于双模态深度学习的钢轨表面缺陷检测方法[J]. 计算机工程与应用, 2023, 59(7): 285−293. doi: 10.3778/j.issn.1002-8331.2209-0364 Zhao H W, Zheng J J, Zhao X X, et al. Rail surface defect method based on bimodal-modal deep learning[J]. Comput Eng Appl, 2023, 59(7): 285−293. doi: 10.3778/j.issn.1002-8331.2209-0364 |
[11] | Hou S T, Dong B, Wang H C, et al. Inspection of surface defects on stay cables using a robot and transfer learning[J]. Autom Constr, 2020, 119: 103382. doi: 10.1016/j.autcon.2020.103382 |
[12] | 李运堂, 谢梦鸣, 王鹏峰, 等. 基于改进YOLOv3算法的斜拉桥拉索表面缺陷检测方法[J]. 传感技术学报, 2021, 34(11): 1509−1517. doi: 10.3969/j.issn.1004-1699.2021.11.014 Li Y T, Xie M M, Wang P F, et al. Defects inspection method for cable surface of cable-stayed bridge based on improved YOLOv3 algorithm[J]. Chin J Sens Actuators, 2021, 34(11): 1509−1517. doi: 10.3969/j.issn.1004-1699.2021.11.014 |
[13] | 孙泽强, 陈炳才, 崔晓博, 等. 融合频域注意力机制和解耦头的YOLOv5带钢表面缺陷检测[J]. 计算机应用, 2023, 43(1): 242−249. doi: 10.11772/j.issn.1001-9081.2021111926 Sun Z Q, Chen B C, Cui X B, et al. Strip steel surface defect detection by YOLOv5 algorithm fusing frequency domain attention mechanism and decoupled head[J]. J Comput Appl, 2023, 43(1): 242−249. doi: 10.11772/j.issn.1001-9081.2021111926 |
[14] | Huang X Y, Liu Z L, Zhang X Y, et al. Surface damage detection for steel wire ropes using deep learning and computer vision techniques[J]. Measurement, 2020, 161: 107843. doi: 10.1016/j.measurement.2020.107843 |
[15] | 张银胜, 杨宇龙, 吉茹, 等. 改进YOLOv5s的风力涡轮机表面缺陷检测[J]. 电子测量与仪器学报, 2023, 37(1): 40−49. doi: 10.13382/j.jemi.B2205857 Zhang Y S, Yang Y L, Ji R, et al. Surface defect detection of wind turbine based on YOLOv5s[J]. J Electron Meas Instrum, 2023, 37(1): 40−49. doi: 10.13382/j.jemi.B2205857 |
[16] | Qiao W T, Liu Q W, Wu X G, et al. Automatic pixel-level pavement crack recognition using a deep feature aggregation segmentation network with a scSE attention mechanism module[J]. Sensors, 2021, 21(9): 2902. doi: 10.3390/s21092902 |
[17] | Li K Y, Wang X F, Ji L J. Application of multi-scale feature fusion and deep learning in detection of steel strip surface defect[C]//Proceedings of 2019 International Conference on Artificial Intelligence and Advanced Manufacturing, 2019: 656–661. https://doi.org/10.1109/AIAM48774.2019.00136. |
[18] | Fan H D, Zhu D Q, Li Y H. An improved yolov5 marine biological object detection algorithm[C]//Proceedings of the 2nd International Conference on Artificial Intelligence and Computer Engineering, 2021: 29–34. https://doi.org/10.1109/ICAICE54393.2021.00014. |
[19] | Strudel R, Garcia R, Laptev I, et al. Segmenter: transformer for semantic segmentation[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision, 2021: 7262–7272. https://doi.org/10.1109/ICCV48922.2021.00717. |
[20] | Zhao Z Y, Yang X X, Zhou Y C, et al. Real-time detection of particleboard surface defects based on improved YOLOv5 target detection[J]. Sci Rep, 2021, 11(1): 21777. doi: 10.1038/s41598-021-01084-x |
[21] | Shelhamer E, Long J, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39(4): 640−651. doi: 10.1109/TPAMI.2016.2572683 |
In recent years, the construction technology of large-span bridges has developed rapidly and its application has increased. As the main form of large-span bridges, the cable-stayed bridge has outstanding advantages such as beautiful appearance, strong seismic resistance, long span distance, low cost, and convenient construction. Therefore, it is widely used in bridges for crossing rivers or seas. As the main load-bearing component, the cable guarantees cable-stayed bridges being serviced safely. The interior of the cable is composed of high-strength steel wires and anti-corrosion coatings while the exterior is mainly protected by polyethylene or high-density polyethylene. Due to long-term exposure to the natural environment and affected by sunlight, wind, rain, and other factors, the protective layer of the cable is extremely easy to be erosion, deformation, cracking, and even peeling, which leads to the failure of the protective function. Furthermore, corrosion media and humid water mist entering the interior of the cable will cause steel wire corrosion and fracture. Therefore, regular cable detection is necessary to ensure bridge safety. Due to the low efficiency, high cost, and poor safety of manual detection of cable surface defects in cable-stayed bridges, existing target detection methods have low accuracy and slow speed, and are easily affected by cable surface dirt interference, resulting in false or missed detections. Therefore, an improved YOLOv5s network is proposed to achieve fast and accurate detection of cable surface defects. Add a TRANS module to the backbone network to obtain more features from a single image and improve defect detection accuracy. In the neck network, GhostBottleneck is used instead of the CSP module, and depthwise separable convolution is used instead of regular convolution to ensure detection accuracy while effectively reducing network parameters and significantly improving detection speed. Introducing the SIOU loss function to solve the problem of mismatch between the real and predicted boxes of small target defects, and improving the convergence speed and stability of the network. Using polyvinyl chloride pipes to simulate cable protection sleeves, constructing a dataset for experiments. The experimental results show that the mAP and FPS of the improved YOLOv5s network reach 94.26% and 68 frames per second, respectively, which are superior to Faster RCNN, YOLOv4, conventional YOLOv5, and other networks, meeting the requirements of surface defect detection accuracy and real-time performance for cable-stayed bridges.
Structure of the conventional YOLOv5s network
SPPF module
Schematic diagram of the prediction box
Structure of the improved YOLOv5s network
TRANS module
Multi-head attention
GhostBottleneck module
Ghost structure
Angle loss
Distance loss
Data collection of cable surface defects
Mosaic data augmentation
Annotation instance
Loss value variation curve
Comparison of different network detection results