基于PP-YOLOE的城市排水管网缺陷检测及应用  

Defect Detection and Application of Urban Drainage Network Based on PP-YOLOE

作　　者：王守志 刘章 王冬 万玉生 高腾 李旭 WANG Shou‑zhi;LIU Zhang;WANG Dong;WAN Yu‑sheng;GAO Teng;LI Xu(North China Municipal Engineering Design&Research Institute Co.Ltd.,Tianjin 300381,China;Service Center for Quality and Safety Management of Construction Projects in Xiong’an New District,Xiong’an 071800,China)

机构地区：[1]中国市政工程华北设计研究总院有限公司,天津300381 [2]雄安新区建设工程质量安全检测服务中心,河北雄安071800

出　　处：《中国给水排水》2024年第18期130-136,共7页China Water & Wastewater

摘　　要：城市排水管网所处环境较为复杂,需要定期进行维护,相关人员需对管道机器人采集的管网CCTV视频逐个进行检查判断。为了节约人力成本,采用深度学习中的目标检测算法(PPYOLOE)智能检测视频中的缺陷信息,并与零样本检测算法(Grounding DINO)进行对比。测试结果表明,PP-YOLOE算法检测的精确率、召回率、准确率分别为1.000、0.875、0.944,大大优于Grounding DINO算法,更适用于排水管网场景。随后,将含有缺陷信息的图片展示在排水管网三维GIS可视化管理平台,便于管理人员直观掌握缺陷情况,辅助提供决策支撑。该项研究成果已在黑龙江某地区的排水管网应用,并取得了较好的效果。Due to the complex environment of urban drainage networks,regular maintenance is required.Relevant personnel need to check and judge each CCTV video of the pipeline network collected by the pipeline robot.In order to save labor costs,the object detection algorithm PP-YOLOE of deep learning is adopted to intelligently detect defect information in videos,and compared with the zero‑sample detection algorithm of Grounding DINO.The test results show that detection precision,recall,and accuracy rate of the PP-YOLOE algorithm are 1.000,0.875,and 0.944,respectively,which are significantly better than the Grounding DINO algorithm and are well suited for drainage network scenarios.Subsequently,images containing defect information are displayed on the three‑dimensional GIS visualization management platform of the drainage network,making it easier for management personnel to intuitively grasp the defect situation and assist in providing decision‑making support.The research results have been applied in the drainage network of a certain area in Heilongjiang Province,and have achieved good results.

关 键 词：管网缺陷检测 CCTV检测 PP-YOLOE算法 Grounding DINO算法 三维GIS 

分 类 号：TU992[建筑科学—市政工程]