基于改进YOLOv8的城市排水管道缺陷检测算法研究  

Enhanced urban drainage pipeline defect detection algorithm based on improved YOLOv8

作　　者：杨帆[1] 刘如飞[2] 刘扬胜 宋佰万 牛冲[1] 来瑞鑫 YANG Fan;LIU Rufei;LIU Yangsheng;SONG Baiwan;NIU Chong;LAI Ruixin(Shandong Geological Survey Institute,Jinan 250003,China;School of Surveying and Spatial Information,Shandong University of Science and Technology,Qingdao 266590,China)

机构地区：[1]山东省地质测绘院,济南250003 [2]山东科技大学测绘与空间信息学院,青岛266590

出　　处：《给水排水》2024年第8期120-125,共6页Water & Wastewater Engineering

摘　　要：排水管道系统在城市管理中起着关键作用,为了实现排水管道缺陷的自动化检测,提出了一种基于改进YOLOv8的排水管道缺陷检测算法。首先针对管道图像亮度不均和网络泛化能力差的问题,采用Zero-DCE亮度增强和图像对比度调整相结合的方法进行数据增强处理。然后通过对YOLOv8算法添加Coordinate Attention注意力机制,增强算法对缺陷位置信息的感知和捕捉能力,以便于算法能够更好的识别排水管道细小缺陷。试验结果表明,相较于原始YOLOv8算法,改进后的算法精确度和召回率分别提升5%和7.9%。与其他三种网络相比,精确度和召回率分别提高了5.5%、7.6%、2.2%和7.9%、4.2%、2%。Drainage pipe system plays a key role in urban management,in order to realize the automated detection of drainage pipe disease.In this paper,we propose a drainage pipe disease detection algorithm based on improved YOLOv8.First for the problem of uneven brightness of pipeline images and poor network generalization ability,data enhancement processing using a combination of Zero-DCE brightness enhancement and image contrast adjustments.Then by adding the Coordinate Attention(CA)attention mechanism to the YOLOv8 algorithm,enhancing the algorithm s ability to perceive and capture disease location information,so that the algorithms can better identify minor drainage pipe defects.The experimental results show that compared to the original YOLOv8 algorithm,the improved algorithm increases precision and recall by 5%and 7.9%respectively.Compared to the other three networks,the precision and recall are improved by 5.5%.7.6%,2.2% and 7.9%,4.2%,2%respectively.

关 键 词：排水管道缺陷 YOLOv8 注意力机制 数据增强 

分 类 号：TU992[建筑科学—市政工程]