排水管网是城市重要的基础设施之一，对保障城市安全运行具有十分重要的作用。为了保证排水管网的完整与通畅，检查员通过闭路电视（CCTV）定期对管道内部进行检测与排查，确定缺陷的种类、位置与等级判定。然而，目前确定缺陷等级普遍采用人工判读方式，效率十分低下。为了实现缺陷等级智能化定量分析，论文研究了排水管道检测现状，构建了基于视觉计算的排水管道缺陷评估模型，在其基础上，开展了以下关键技术的研究：

（1）针对标准霍夫梯度圆检测（21HT）在排水管道检测图像上的局限性，提出了基于约束型霍夫变换（RHGT）的管道特征提取算法，予以候选圆一定的约束策略，并通过寻优策略保证结果最优。通过实验验证了RHGT算法能够过滤冗余的假圆数据，最终确定的管道截面圆有效拟合了缺陷相对位置。实验结果表明在缺陷评估模型中部署RHGT算法相比21HT算法，平均准确率提高了22.15%。

（2）针对缺陷特征提取不充分和自适应性差问题，提出了基于边缘检测的缺陷特征提取方法。一方面，基于整体嵌套边缘检测（HED）模型自适应提取全局语义级的缺陷轮廓，通过图像形态学操作和Canny算子进一步提取显著性边缘，进而计算缺陷面积；另一方面，基于YOLO-OTSU模型分割精细缺陷轮廓。实验结果表明，所提出的边缘检测缺陷特征提取方法具有自适应性，无需手动调整阈值，并且，在多缺陷样本上特征提取的缺陷轮廓更加完整。

（3）针对缺陷等级人工判定方法存在的主观不确定性问题，提出了基于效用函数的缺陷等级评估方法。基于效用理论，分析我国排水管道评估规程，建立了图像特征到缺陷效用值再到缺陷等级间联系。依据多项式回归法为10种缺陷构建了不同的效用函数，拟合优度R² 从0.918到0.99，表明效用函数与人工评判的结果保持一致，实现了评估方法从主观语言描述到客观准确计算的转变。

部署上述三个关键算法到管道缺陷评估模型，在自主构建的松柏路数据集和Level-Sewer10两个数据集上进行了实验。实验结果表明，与人工检测报告对比，所提缺陷评估模型的平均绝对偏差为2.008%，平均准确率为86.73%。对视频数据的实验分析结果表明，所构建的模型能正确检测出视频中存在的缺陷并输出相应的缺陷等级，具有实际应用价值。

Drainage pipe network is one of the important infrastructures of the city, which plays a vital role in ensuring the safe operation of the city. To ensure the integrity and smoothness of the drainage network, inspectors regularly investigate the inside of the pipeline through closed-circuit television (CCTV) to determine the type, location and grade of defects. However, at present, manual interpretation is generally used to determine the defect level, which leads to low efficiency. To realize the intelligent quantitative analysis of defect level, this paper studies the current situation of drainage pipeline detection. On the basis of building a drainage pipeline defect assessment model based on visual computing, the following key algorithms are studied.

(1) Aiming at the limitation of standard Hough gradient circle detection (21HT) in the inspected image of drainage pipelines, this paper proposed the restricted Hough gradient transform (RHGT) to extract the pipeline feature. In the RHGT algorithm, certain constraint strategies are given to the candidate circles to ensure that the results of the algorithm are optimized. It is verified by experiments that the RHGT algorithm can effectively filter the redundant candidate false circles, and the final determined pipe section circle can effectively fit the relative position of the defect. The experimental results show that deploying the RHGT algorithm in the defect evaluation model improves the average accuracy by 22.15% compared with the 21HT algorithm.

(2) Aiming at the problem of insufficient defect feature extraction and poor adaptability, a defect feature extraction method based on edge detection is proposed. On the one hand, the global semantic-level defect contour is adaptively extracted based on the Holly-Nested edge detection (HED) model, and the salient edges are further extracted through image morphological operations and Canny operator. On the other hand, the YOLO-OTSU model segments detailed defect contours. The experimental results show that the proposed edge detection defect feature extraction method is self-adaptive and does not need to manually adjust the threshold value. Furthermore, the defect contours extracted by the feature extraction on multi-defect samples are more integral.

(3) Aiming at the subjective uncertainty of the manual determination of defect level, an automated defect level evaluation method based on utility function is proposed. Based on the utility theory, this paper analyzes the evaluation procedures of drainage pipelines of China, and establishes the relationship between the image features, the utility value of the defect, and the defect level. According to the polynomial regression method, different utility functions are constructed for 10 kinds of defects. The coefficient of determination R² ranges from 0.918 to 0.99, indicating that the utility function is consistent with the results of manual evaluation. Hence, the evaluation method has been transformed from subjective language description to objective and accurate calculation.

The above three key algorithms are deployed to the pipeline defect evaluation model, and experiments are carried out on the self-built Songbai dataset and Level-Sewer10 dataset. The experimental results of images show that the average absolute deviation of the proposed defect evaluation model is 2.008%, and the average accuracy rate is 86.73% compared with the manual inspection reports. The experimental analysis results of videos show that the proposed model can correctly detect the defects in the video and output the corresponding defect level, which has practical application value.

[1]Guo W, Soibelman L, Jr J. Automated defect detection for sewer pipeline inspection and condition assessment[J]. Automation in Construction, 2009, 18(5):587-596.

[2]陈伟珂, 丁聿. 城市排水管网脆弱性变权评估研究[J]. 水资源与水工程学报, 2019, 30(02):139-146.

[3]中华人民共和国住房和城乡建设部. 中国城乡建设统计年鉴 2020[M]. 北京: 中国统计出版社. 2020.09.

[4]OECD. Financing water supply, sanitation and flood protection: challenges in EU member states and policy options[R]. Paris, France: OECD Publishing. 2020.

[5]Moradi S, Zayed T, Golkhoo F. Review on Computer Aided Sewer Pipeline Defect Detection and Condition Assessment[J]. Infrastructures, 2019, 4(1):1-15.

[6]Rizzo, P. Water and Wastewater Pipe Nondestructive Evaluation and Health Monitoring: A Review[J]. Advances in Civil Engineering, 2010, 2010:1-14.

[7]王俊岭, 邓玉莲, 李英等．排水管道检测与缺陷识别技术综述[J]. 科学技术与工程, 2020, 20(33):13520-13528.

[8]赵永强, 饶元, 董士鹏等. 深度学习目标检测方法综述[J]. 中国图像图形学报, 2020, 25(4):629-654.

[9]Ye X, Zuo J, Li R, et al. Diagnosis of sewer pipe defects on image recognition of multi-features and support vector machine in a southern Chinese city[J]. Frontiers of Environmental Science & Engineering, 2019, 13(2):29-41.

[10]Cheng J C P, Wang M. Automated detection of sewer pipe defects in closed-circuit television images using deep learning techniques[J]. Automation in Construction, 2018, 95(NOV):155-171.

[11]Li D, Cong A, Guo S. Sewer damage detection from imbalanced CCTV inspection data using deep convolutional neural networks with hierarchical classification[J]. Automation in Construction, 2019, 101(MAY):199-208.

[12]Makar, J.M. Diagnostic techniques for sewer systems[J]. Journal of Infrastructure Systems, 1999, 5(2):69–78.

[13]Su, T.C., Yang, M.D., Wu, T.C., Lin, J.Y. Morphological segmentation based on edge detection for sewer pipe defects on cctv images[J]. Expert Systems with Applications, 2011, 38(10): 13094–13114.

[14]Halfawy, M., Hengmeechai, J. Efficient algorithm for crack detection in sewer images from closed-circuit television inspections[J]. Journal of Infrastructure Systems, 2014, 20(2):1-12.

[15]黄玉龙. 基于视频图像的管道裂纹缺陷检测方法研究[D]. 西安: 西安理工大学，2018.

[16]Hawari A, Alamin M, Alkadour F, et al. Automated defect detection tool for closed circuit television (cctv) inspected sewer pipelines[J]. Automation in Construction, 2018, 89(MAY):99-109.

[17]Burges C . A Tutorial on Support Vector Machines for Pattern Recognition[J]. Data Mining and Knowledge Discovery, 1998, 2(2):121-167.

[18]JC A, Gl A, Rm B, et al. A comprehensive survey on support vector machine classification: Applications, challenges and trends-ScienceDirect[J]. Neurocomputing, 2020, 408(SEPT):189-215.

[19]Yang, M.D., Su, T.C. Automated diagnosis of sewer pipe defects based on machine learning approaches[J]. Expert Systems with Applications, 2008, 35(3):1327-1337.

[20]Halfawy M R, Hengmeechai J. Automated defect detection in sewer closed circuit television images using histograms of oriented gradients and support vector machine[J]. Automation in Construction, 2014, 38(MAR):1-13.

[21]Wu, S., Zhong, S., Liu, Y. Deep residual learning for im- age steganalysis[J]. Multimedia Tools and Applications, 2018, 77(FEB):10437–10453.

[22]MA Gutierrez-Mondragon, Garcia-Gasulla D, Alvarez-Napagao S, et al. Obstruction level detection of sewer videos using convolutional neural networks[J/OL]. 2020.

[23]Haurum J B, Moeslund T B. A Survey on Image-Based Automation of CCTV and SSET Sewer Inspections[J]. Automation in construction, 2020, 111(103061):1-19.

[24]Water Research Centre. Sewerage rehabilitation manual (4th Ed.) [S], U.K.: Water Research Centre, U.K. 2004.

[25]NZPIM. New Zealand pipe inspection manual[S]. New Zealand: New Zealand Water and Waste Water Association, 2006.

[26]PACP, Pipeline assessment and certification program[S]. U.S.A.: North American Society of Sewer Service Companies (NASSCO), 2019.

[27]Wong, K., Allen, R.J. Utility Training Institute., Hong Kong conduit condition evaluation codes 2009 (HKCCEC2009): code of practice for conduit condition evaluation using CCTV in Hong Kong, 4th ed[S]. Hong Kong: Advanced Technovation Ltd, 2009.

[28]CJJ181-2012. 城镇排水管道检测与评估技术规程[S]. 北京: 中国建筑工业出版社，2012.

[29]中华人民共和国住房和城乡建设部. 住房城乡建设部关于发布行业标准《城镇排水管道检测与评估技术规程》的公告[EB]. 2012. (http://www.mohurd.gov.cn/gongkai/fdzdgknr/tzgg/201210/20121023_211724.html)

[30]胡晓庆. 宜兴市排水管道缺陷诊断量化方案及效果评价[D]. 北京: 清华大学, 2015.

[31]Wang M, Luo H, Cheng JCP. Towards an automated condition assessment framework of underground sewer pipes based on closed-circuit television (CCTV) images[J]. Tunnelling and Underground Space Technology, 2021, 110:1-20.

[32]殷德雨. 多传感器融合地下管道缺陷智能检测关键技术研究[D].武汉:武汉大学,2020.

[33]庄福振, 罗平, 何清, 史忠植. 迁移学习研究进展[J]. 软件学报, 2015, 26(1): 26-39.

[34]J. Redmon, S. Divvala, R. Girshick and A. Farhadi. You Only Look Once: Unified, Real-Time Object Detection[C] // IEEE. IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Las Vegas, NV, USA, IEEE, 2016:779-788.

[35]Waibel P, Matthes J, Grll L. Constrained Ellipse Fitting with Center on a Line[J]. Journal of Mathematical Imaging and Vision, 2015(53):364-382.

[36]Yuen H K, Princen J, Illingworth J, et al. Comparative study of Hough Transform methods for circle finding[J]. Image & Vision Computing, 1989, 8(1):71-77.

[37]Wang M, Luo H, Cheng J. Severity Assessment of Sewer Pipe Defects in Closed-Circuit Television (CCTV) Images Using Computer Vision Techniques[C] // ASCE. Construction Research Congress 2020: Infrastructure Systems and Sustainability. Baltimore: ASCE Library, 2020:942-950.

[38]Hough P. Method and Means for Recognizing Complex Patterns[J/OL]. 美国能源部科技信息办公室, 1962. (https://www.osti.gov/biblio/4746348)

[39]刘喆. 基于图像分析的管道缺陷特征提取方法研究[D]. 沈阳:东北大学, 2014.

[40]Marr D, Hildreth E. Theory of Edge Detection[J]. Proceedings of the Royal Society of London. Series B. Biological Science, 1980, 207:187-217.

[41]张岩, 赵玉涛, 李光旭,等. 图像的边缘检测技术研究进展[J]. 计算机与数字工程, 2020, 48(5):1176-1184.

[42]Canny, J. A computational approach to edge detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1986, PAMI-8(6):679-698.

[43]Xie S, Tu Z. Holistically-Nested Edge Detection[C]// IEEE. 2015 IEEE International Conference on Computer Vision (ICCV). Santiago, Chile: IEEE Xplore, 2016:1395-1403.

[44]Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-Scale Image Recognition[C] // Computational and Biological Learning Society. 2015 International Conference on Learning Representations. San Diego, CA, USA, dblp Computer Science, 2015:1-14.

[45]焦安波, 何淼, 罗海波. 一种改进的HED网络及其在边缘检测中的应用[J]. 红外技术, 2019, 41(1):72-77.

[46]Otsu N. A Threshold Selection Method from Gray-Level Histograms[J]. IEEE Transactions on Systems Man & Cybernetics, 2007, 9(1):62-66.

[47]Suzuki S, Be K. Topological structural analysis of digitized binary images by border following[J]. Computer Vision Graphics & Image Processing, 1985, 30(1):32-46.

[48]赵延军, 王晓鸣, 李寓, 薛文碧. 工程项目的效用决策方法[J]. 西安科技大学学报, 2009(5):543-548.

[49]苏为华. 多指标综合评价理论与方法问题研究[D]. 厦门:厦门大学, 2000.

[50]罗玮. 基于计划行为和多属性效用理论的PC桥梁可持续加固策略[D]. 北京: 北京交通大学,2021.

[51]Kaddoura K, Zayed T, Hawari A H. Multiattribute Utility Theory Deployment in Sewer Defects Assessment[J]. Journal of Computing in Civil Engineering, 2018, 32(2):1-13.

[52]Alvarez J, Mercado P E. Online Inference of the Dynamic Security Level of Power Systems Using Fuzzy Techniques[J]. IEEE Transactions on Power Systems, 2007, 22(2):717-726.

[53]Haurum, Joakim Bruslund and Moeslund, Thomas B. Sewer-ML: A Multi-Label Sewer Defect Classification Dataset and Benchmark[C]. // Proceedings of the IEEE. CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021.