基于改进YOLOv5的地下排水管道缺陷检测算法  

Defect detection algorithm of underground drainage pipe based on improved YOLOv5

作　　者：吴谦 孙丙宇[2] 房永峰 WU Qian;SUN Bingyu;FANG Yongfeng(School of Mechanical and Electrical Engineering,Anhui Jianzhu University,Hefei Anhui 230601,China;Hefei Institute of Physical Science,Chinese Academy of Sciences,Hefei Anhui 230026,China;Graduate School of University of Science and Technology of China,Science Island Branch,Hefei Anhui 230026,China)

机构地区：[1]安徽建筑大学电子与信息工程学院,安徽合肥230601 [2]中国科学院合肥物质科学研究院,安徽合肥230026 [3]中国科学技术大学研究生院科学岛分院,安徽合肥230026

出　　处：《阜阳师范大学学报（自然科学版）》2024年第3期21-27,共7页Journal of Fuyang Normal University:Natural Science

基　　金：中国科学院合肥物质科学研究院院长基金重点支持项目(YZJJZX202013)。

摘　　要：针对地下排水管道缺陷检测存在效率低和精度差等问题,本文提出了一种基于改进的YOLOv5模型的地下排水管道缺陷识别方法。首先,为了提高检测精度,引入卷积块注意力机制模块(Convolutional Block Attention Module,CBAM)解决了部分缺陷过小的问题,同时设计了一种新的空间金字塔池化模块,解决不同类型缺陷大小、形状和颜色相差较大的问题。其次,在特征融合层引入加权双向特征金字塔网络结构(bi-directional feature pyramid network structure,BiFPN)以充分利用不同尺度的特征,提高了特征融合能力。最后,针对图片模糊的问题,对原始图像进行了直方图均衡化、锐化处理。实验结果表明,改进后的算法能很好对障碍物、错口、裂缝、树根和错口等5种缺陷进行识别,平均精度(Mean Average Precision,mAP)相比原始YOLOv5算法提升了2.5%。Aiming at the problems of low efficiency and poor accuracy in the detection of underground drainage pipe defects,this paper proposes a defect identification method for underground drainage pipes based on the improved YOLOv5 model.Firstly,in order to improve the detection accuracy,the convolutional block attention mechanism module(CBAM)is introduced to solve the problem that some defects are too small,and a new spatial pyramid pooling module is designed to solve the problem of large differences in size,shape and color of different types of defects.Secondly,a weighted bidirectional feature pyramid network structure(BiFPN)is introduced in the feature fusion layer to make full use of features of different scales,which improves the feature fusion ability.Finally,aiming at the problem of blurred pictures,the histogram equalization and sharpening of the original image is carried out.The experimental results show that the proposed algorithm can well identify five kinds of defects such as obstacles,wrong mouths,cracks,tree roots and wrong mouths,and the mean average precision is 2.5%higher than the original YOLOv5 algorithm.

关 键 词：目标检测 管道缺陷检测 YOLOv5 缺陷识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]