| 基于深度学习的排水管道缺陷内窥检测智能识别系统研究 |
| |
| 引用本文： | 钟洪德.基于深度学习的排水管道缺陷内窥检测智能识别系统研究[J].城市勘测,2022(1):165-170. |
| |
| 作者姓名： | 钟洪德 |
| |
| 作者单位： | 福州市勘测院,福建 福州 350108 |
| |
| 摘 要： | 目前国内各城市已普遍采用管道机器人深入管道内部摄取视频影像,有效获取到可供管道缺陷检测的一手资料,但缺陷识别大部分依靠人工目视识别,耗时耗力,生产周期长。利用福州市勘测院多年累积的管道检测数据,基于Pytorch深度学习框架、建立了排水管道缺陷内窥检测智能识别系统,包括:数据预处理,残差神经网络设计与训练、系统集成等。重点实现了三级组合识别模型建构(二分类,类型识别,等级识别),解决了系统准确度等技术难题。经生产实践表明:模型准确率高,可有效提高管道健康状况检查质量和效率。
|
| 关 键 词： | 管道缺陷识别 深度学习 卷积神经网络 |
Research on Intelligent Recognition System of Drainage Pipeline Defect Endoscopic Detection Based on Deep Learning |
| |
| Zhong Hongde.Research on Intelligent Recognition System of Drainage Pipeline Defect Endoscopic Detection Based on Deep Learning[J].Urban Geotechnical Investigation & Surveying,2022(1):165-170. |
| |
| Authors: | Zhong Hongde |
| |
| Affiliation: | (Fuzhou Investigation and Survey Institute,Fuzhou 350108,China) |
| |
| Abstract: | At present,pipeline robot has been widely used to capture video images in the pipeline to effectively obtain the first-hand information for pipeline defect detection.However,defect identification mostly relies on manual visual identification,which is time-consuming and labor-consuming,and has a long production cycle.Based on the pipeline inspection data accumulated by Fuzhou Investigation and Survey Institute for many years,an intelligent recognition system for the defects of drainage pipeline is established based on deep learning under the framework of Pytorch,including data preprocessing,residual neural network design and training,system integration and soon.This paper focuses on the realization of three-level combination recognition model construction(two classification,type recognition,grade recognition),and solves the technical problems such as system accuracy.The production practice shows that the model has high accuracy and can effectively improve the quality and efficiency of pipeline health inspection. |
| |
| Keywords: | pipeline defect identification deep learning convolutional neural network |
| 本文献已被 维普 万方数据 等数据库收录！ |
|
