改进的YOLOv5积水检测算法
Improved water accumulation detection algorithm based on YOLOv5
杨庆江 1信威 1马仲甜 1李钊枢 2吕丛生 2赵斌2
作者信息
- 1. 黑龙江科技大学,电子与信息工程学院,哈尔滨 150022
- 2. 国网黑龙江省电力有限公司 黑河供电公司,黑龙江 黑河 164300
- 折叠
摘要
为了提升YOLOv5 算法的积水检测精准度,提出一种改进YOLOv5 网络结构的积水检测算法.通过解耦头代替初始YOLOv5 的耦合头,使分类任务与回归任务两者互不干扰,在网络训练到300 轮次左右时表现出了良好的收敛效果;利用CIoU损失函数代替GIoU,降低积水检测的误差,提高积水检测精准度.结果表明,相比初始YOLOv5 算法的精准度、召回率、mAP1和mAP2分别提升了1.7%、1.8%、1%和0.5%.文中的积水检测算法比初始算法减少了训练损失,收敛速度更快,精准度更高,可为积水区域的检测提供算法支撑.
Abstract
This paper aims to improve the accuracy of water accumulation detection based on YOLOv5 algorithm,and proposes an improved water accumulation detection algorithm based on YOLOv5 network structure.By replacing the initial YOLOv5 coupling head with a decoupling head,the classifica-tion task and regression task are not interfered with each other,and a good convergence effect is shown after about 300 rounds of network training.The study improves the accuracy of water accumulation detec-tion by using CIOU loss function instead of GIOU to reduce the error of water accumulation detection.The results show that compared with the initial YOLOv5 algorithm,the accuracy,recall rate,mAP1,and mAP2 are increased by 1.7%,1.8%,1% and 0.5%,respectively.The water accumulation detection algo-rithm in this paper reduces the training loss compared with the initial algorithm,has a faster convergence speed,and higher accuracy,as which can provide the algorithm support for the detection of water accu-mulation areas.
关键词
积水检测/YOLOv5/解耦头/CIoUKey words
water accumulation detection/YOLOv5/decoupling head/CIoU引用本文复制引用
基金项目
黑龙江省"揭榜挂帅"科技攻关项目(2021ZXJ02A02)
国家电网科学技术项目(SGHLHH00SZJS2310193)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据