电力工程设计中铁塔设计图纸的自动识别与信息提取是一项急需解决的问题。该文提出一种基于深度学习和光学字符识别(Optical Character Recognition，OCR)技术的铁塔设计图纸智能识别系统。该系统由分段结构识别、文本识别和关键信息提取3个主要模块组成。分段结构识别模块采用改进的U-Net卷积神经网络模型;文本识别模块基于Tesseract 4。0 进行优化，提高字符识别准确率;关键信息提取模块则使用基于规则的解析引擎，从识别出的分段结构和文本中抽取关键信息。实验结果表明，该系统在铁塔图纸识别的准确性、泛化性和效率方面均达到较高水平塔形结构识别F1 值为 96。35%，字符识别准确率为 99。10%。该系统可有效支持电力工程设计和管理的数字化、智能化转型，具有广阔的应用前景。
Automatic identification and information extraction of tower design drawings in power engineering design is an urgent problem to be solved urgently.This paper proposes an intelligent recognition system for tower design drawings based on deep learning and optical character recognition(OCR)technology.The system consists of three main modules:segmented structure recognition,text recognition and key information extraction.The segmented structure recognition module adopts an improved U-Net convolutional neural network model;the text recognition module is optimized based on Tesseract 4.0,which improves the accuracy of character recognition.The key information extraction module uses a rule-based parsing engine to extract key information from the identified segmentation structures and texts.Experimental results show that the system achieves a higher level tower structure recognition F1 value of 96.35%and a character recognition accuracy of 99.10%in terms of accuracy,generalization and efficiency in tower drawing recognition.The system can effectively support the digital and intelligent transformation of power engineering design and management,and has broad application prospects.
tower drawingdeep learningoptical character recognition(OCR)key information extractionU-NetTesseract
万方数据
维普
