摘要
快速城市化进程与频发的极端降雨事件相叠加,导致城市内涝加剧,积水灾害频发。快速准确地获得道路积水特征数据是解决城市雨洪灾害问题的迫切需求。在已有基于图像识别进行道路积水监测的研究基础上,提出了一种基于视频影像的道路积水实时监测模型,利用YOLOv5s深度学习算法对道路积水图像进行积水区域的识别,利用测量公式获得识别区域的积水面积,并且采用该模型对某大学校园内一次雨后积水进行应用研究,并与SSD和Faster R-CNN算法相对比。结果表明,该模型能满足道路积水的实时监测任务要求,道路积水识别精度均值mAP为96.06%,积水面积提取平均准确率为93.6%,模型的整体性能优于SSD和Faster R-CNN算法。
The superposition of rapid urbanization and frequent extreme rainfall events has led to intensified urban waterlogging and frequent waterlogging disasters.It is an urgent need to quickly and accurately obtain road water feature data to solve the problem of urban rain and flood disasters.Based on the existing research on road water monitoring based on image recognition,a real-time road water monitoring model based on video images is proposed.The YOLOv5s deep learning algorithm is used to identify the water accumulation area of the road water accumulation image,and the measurement formula is used to obtain the water accumulation area of the recognition area.This model is used to study the water accumulation after a rain in a university campus,and compared with SSD and Faster R-CNN algorithm.The results are as follows.The model can meet the real-time monitoring task requirements of road water accumulation.The average mAP of road accumulation recognition accuracy is 96.06%,and the average accuracy of water accumulation area extraction is 93.6%.The overall performance of the model is better than SSD and Faster R-CNN algorithms.
作者
胡昊
李擎
马鑫
陈军朋
孙爽
徐鹏
HU Hao;LI Qing;MA Xin;CHEN Junpeng;SUN Shuang;XU Peng(Yellow River Conservancy Technical Institute,Kaifeng 475004,China;North China University of Water Resources and Electric Power,Zhengzhou 450045,China;Henan Engineering Research Center of Project operation and ecological security for Inter-basin regional water diversion project,Kaifeng 475004,China)
出处
《华北水利水电大学学报（自然科学版）》
北大核心
2023年第1期62-70,共9页
Journal of North China University of Water Resources and Electric Power：Natural Science Edition
基金
河南省重点研发与推广专项(222102320134)
河南省高等学校重点科研项目(22A570006)
河南省高等学校青年骨干教师培养计划项目(2019GGJS105)
开封市重点研发专项(22ZDYF007)。
关键词
视频影像
YOLOv5s
智慧水利
积水监测
目标检测
video image
YOLOv5s
smart water conservancy
water accumulation monitoring
target detection
