基于U-Net的电气二次图纸文字信息自动化检测与识别  

Automated Detection and Recognition of Text Information in Electrical Secondary Drawings Based on U-Net

作　　者：王磊 张建行 何进锋 付同福 WANG Lei;ZHANG Jianhang;HE Jinfeng;FU Tongfu(Liupanshui Power Supply Bureau of Guizhou Power Grid Co.,Ltd.,Liupanshui 553000,China)

机构地区：[1]贵州电网有限责任公司六盘水供电局,六盘水553000

出　　处：《自动化与仪表》2024年第11期141-145,共5页Automation & Instrumentation

基　　金：贵州电网有限责任公司科技项目(060200KK52220001)。

摘　　要：随着工业自动化和智能化的发展,传统的手工查找和读取电气二次图纸的方式已经无法满足现代工程实践的需求,深度学习文字检测方法正应用于电气二次图纸文字信息的自动化检测与识别。针对电气二次图纸风格迥异、文字分布不均的情况,提出使用U-Net深度学习网络完成电气二次图纸文字区域分割,提高自动化处理电气图纸中文字信息的效率和准确性。通过大量数据集对模型进行训练,得到了一个准确且鲁棒的模型。通过实验对比,验证了所提U-Net深度学习方法进行文字自动化检测的有效性及整体算法对于电气二次图纸识别的鲁棒性。With the development of industrial automation and intelligence,the traditional way of manually finding and reading electrical secondary drawings can no longer meet the needs of modern engineering practice,and deep learning text detection methods are being applied to the automatic detection and recognition of text information of electrical secondary drawings.In view of the different styles and uneven text distribution of electrical secondary drawings,the U-Net deep learning network was proposed to complete the text area segmentation of electrical secondary drawings to improve the efficiency and accuracy of automatic processing of Chinese information of electrical drawings.The model was trained on a large number of datasets,and an accurate and robust model was obtained.Through experimental comparison,the effectiveness of the proposed U-Net deep learning method for automatic text detection and the robustness of the overall algorithm for electrical quadratic drawing recognition are verified.

关 键 词：电气二次图纸 深度学习 自动化检测 U-Net 图像分割 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]