基于YOLOv7算法的易涝区积水快速识别方法研究  

A Method for Rapid Identification of Waterlogging in Flood‑prone Area Based on YOLOv7

作　　者：金晓飞 郭帅 黄琼 袁伟 JIN Xiao‑fei;GUO Shuai;HUANG Qiong;YUAN Wei(School of Civil Engineering,Hefei University of Technology,Hefei 230009,China;College of Intelligent Manufacturing,Anhui Vocational and Technical College,Hefei 230011,China;Hefei Surveying and Mapping Design Institute Co.Ltd.,Hefei 230061,China)

机构地区：[1]合肥工业大学土木与水利工程学院,安徽合肥230009 [2]安徽省职业技术学院智能制造学院,安徽合肥230011 [3]合肥市测绘设计研究院有限公司,安徽合肥230061

出　　处：《中国给水排水》2024年第9期123-128,共6页China Water & Wastewater

基　　金：安徽省重点研发计划项目(202104i07020012)。

摘　　要：为了能够高效便捷地监测识别城市易涝区积水情况,利用深度学习技术,提出了一种基于YOLOv7算法的快速识别积水方法。采用传统数据增强和Mosaic数据增强方法对训练集图像进行扩充,构建了YOLOv7积水检测模型,并与其他主流目标检测模型Faster R-CNN和YOLOv5m进行了对比分析。结果表明,YOLOv7模型取得了最好的效果,其精度、召回率、平均精度、F_(1)分数分别达到了92.9%、83.4%、88.8%和87.9%,且单张图片推理时间仅约为0.025 s。该方法在城市内涝积水识别与预警方面具有良好的应用前景。This paper proposed a method for rapid identification of waterlogging based on YOLOv7 algorithm by using deep learning technology,so as to monitor and identify waterlogging in flood‑prone area efficiently and conveniently.The traditional and Mosaic data augmentation methods were used to expand the training set images,and the YOLOv7 waterlogging detection model was established.Then,the model was compared with other mainstream object detection models(Faster R‑CNN and YOLOv5m).The YOLOv7 model achieved the best performance.Its precision,recall,average precision and F_(1)score reached 92.9%,83.4%,88.8%and 87.9%respectively,and the inference time of a single image was only about 0.025 s.This method demonstrates a good application prospect in the identification and early warning of urban waterlogging.

关 键 词：城市内涝 积水识别 目标检测 YOLOv7算法 深度学习 

分 类 号：TU992[建筑科学—市政工程]