基于深度学习的排水管道缺陷检测算法对比研究  

Comparative Study on Deep Learning-based Algorithms for Drainage Pipeline Defect Detection

作　　者：黄琼[1] 郑耀先 宋文博 潘刚[2] 郭帅 HUANG Qiong;ZHENG Yaoxian;SONG Wenbo;PAN Gang;GUO Shuai

机构地区：[1]安徽职业技术学院智能制造学院,安徽合肥230011 [2]天津大学人工智能计算学部,天津300072 [3]合肥工业大学土木与水利工程学院,安徽合肥230009

出　　处：《安徽职业技术学院学报》2024年第2期15-19,25,共6页Journal of Anhui Vocational & Technical College

摘　　要：管道的缺陷检测是城镇排水管道检修和维护的前提,通过准确的缺陷检测技术有利于推动城市管道的建设工程质量。相较于传统的缺陷检测方法,深度学习算法能提高缺陷检测的准确性和可靠性。研究基于管道实测缺陷数据,使用面向排水管道缺陷检测的图像语义分割算法和基于异常检测和半监督学习的排水管道缺陷多分类算法,以检测准确率、精确率和召回率为评估指标,分别检测错口、破裂和支管暗接三类管道缺陷,对比分析两种算法对管道缺陷的检测性能。研究发现,面向排水管道缺陷检测的图像语义分割算法是较为理想的管道缺陷检测模型。The detection of Pipeline defects is the prerequisite for the overhaul and maintenance of urban drainage pipelines.Accurate defect detection technology is conducive to promoting the quality of urban pipeline construction projects.Compared with traditional defect detection methods,deep learning algorithms can improve the accuracy and reliability of defect detection.Based on the measured defect data of pipelines,this study employs the image semantic segmentation algorithm for the detection of drainage pipeline defects and the multi-classification algorithm for the detection of drainage pipeline defects based on anomaly detection and semi-supervised learning.The detection performance of the two algorithms for pipeline defects is compared and analyzed taking the detection accuracy,precision and recall rate of three types of pipeline defects,misalignment,rupture and concealed connections of branch pipes,as the evaluation indicators.It is found that the image semantic segmentation algorithm for drainage pipeline defect detection is an ideal pipeline defect detection model.

关 键 词：深度学习 排水管道 缺陷检测 

分 类 号：TU992.0[建筑科学—市政工程]