摘要
随着现代化进程的不断推进,城市排水系统作为城市基础设施的重要组成部分,其规模和复杂程度不断增加。传统管道缺陷识别依赖于人工判读,费时费力且生产周期长。因此,开展排水管道缺陷智能识别系统研究具有重要的现实意义。本文针对工程项目中的实际需求,提出并实现了一套基于深度学习的复杂城市排水管道缺陷智能识别系统。通过管道缺陷数据库建立、多尺度缺陷检测模型构建、样本增量学习和迁移学习、基于专家系统的评估模型和帧间差分算法等关键技术,实现了缺陷识别的智能化和自动化。经生产实践验证,缺陷智能识别系统精确率达到78.04%,召回率达到84.59%,可有效提高管道缺陷检测效率。
With the continuous advancement of modernization,urban drainage system,as an important part of urban infrastructure,is increasing in scale and complexity.Traditional pipeline defect identification relies on manual interpretation,which is time-consuming,laborious,and has a long production cycle.Therefore,conducting research on intelligent recognition systems for drainage pipeline defects has important practical significance.This article proposes and implements an intelligent recognition system for complex urban drainage pipeline defects based on deep learning,targeting practical needs in engineering projects.Through the pipeline defect database,multiscale defect detection model construction,sample incremental learning and transfer learning,evaluation model based on expert system and inter-frame difference algorithm and other key technologies,the intelligent and automatic defect identification is realized.The precision rate of the intelligent defect identification system is 78.04%and the recall rate is 84.59%,which can effectively improve the efficiency of pipeline defect detection.
作者
许铁林
廖立国
周昌林
XU Tielin;LIAO Liguo;ZHOU Changlin(Chengdu Surveying Geotechnical Research Institute Co.,Ltd.,MCC,Chengdu 610023,China)
出处
《测绘通报》
CSCD
北大核心
2024年第S02期37-41,共5页
Bulletin of Surveying and Mapping
基金
2022年度四川省住房城乡建设领域科技创新立项课题(SCJSKJ2022-14)
关键词
深度学习
排水管道
缺陷识别
增量学习
deep learning
drainage pipeline
defect identification
incremental learning
