北京测绘2025，Vol.39Issue(11)：1648-1653,6.DOI:10.19580/j.cnki.1007-3000.2025.11.014
基于深度学习的规划道路数据智能化质检方法
Intelligent quality inspection of planned road data based on deep learning
摘要
Abstract
As an important outcome of engineering surveying,planned road data serves as the fundamental data foundation and guarantee for urban planning and approval.Due to the phased construction of the planned road database,many data results cannot be entered into the database in a timely and accurate manner,leading to inconsistencies between the database data and road design data.Furthermore,the large amount of planned road data makes manual quality inspection time-consuming,labor-intensive,and difficult to ensure accuracy.Therefore,this paper proposed an intelligent quality inspection scheme for planned road data based on deep learning technologies such as computer vision and large models.The intelligent quality inspection solution utilized the convolutional networks for biomedical image segmentation(U-Net)convolutional net-work model to detect table areas in road design images.It then combined expert knowledge to complete the analysis of the table structure.Finally,based on paddle optical character recognition(PaddleOCR),cell character recognition was per-formed to achieve the structuring and informatization of the table images.In addition,this paper innovatively introduced large model technologies to correct textual similarity errors,further improving recognition accuracy.Through implementa-tion and validation,the intelligent quality inspection solution saves more than 90%of inspection time,with a misidentifica-tion rate of only 0.27%,greatly enhancing both efficiency and accuracy.This demonstrates the effectiveness of this method in planned road quality inspection.关键词
规划道路数据质检/计算机视觉/表格检测/大语言模型/字形相似性Key words
planned road data quality inspection/computer vision/table detection/large language model/character similarity分类
天文与地球科学引用本文复制引用
刘世凡,邢晨,董承玮,马金荣,金鹏,李晓燕..基于深度学习的规划道路数据智能化质检方法[J].北京测绘,2025,39(11):1648-1653,6.基金项目
科技部创新工作方法专项(2020IM020500) （2020IM020500）
