电子器件2025，Vol.48Issue(2)：339-345,7.DOI:10.3969/j.issn.1005-9490.2025.02.016
基于OCR和点位配准的电表检定算法
An Electric Meter Appearance Verification Algorithm with OCR and Point Matching
安泰 1赵森 2谷依田 3程大伟 2张涛3
作者信息
- 1. 国电南瑞科技股份有限公司南京用电技术分公司,江苏 南京 211106
- 2. 北京南瑞捷鸿科技有限公司,北京 100193
- 3. 东南大学仪器科学与工程学院,江苏 南京 210096
- 折叠
摘要
Abstract
The meter appearance verification system is important to the evaluation of meter production.In view of the low recognition ac-curacy of the character defect detection system with dense text distribution,the adaptive multi-scale module is used to improve the de-tection accuracy and segmentation quality of the algorithm.The STN attention module for improving the robustness of scene text recogni-tion under the influence of tilt and perspective is used to warp the text image into the most appropriate angle as the input image for neu-ral network.A point matching algorithm is proposed to find the character defects accurately and decrease the misjudgment of the algo-rithm.Experiments show that accuracy improvements of 6.1%and 4.3%are achieved in detection algorithm and recognition algorithm respectively.Point matching algorithm achieves the F-score of 97.8%.With duration of 0.8 s per electric meter,the verification algo-rithm meets the real-time requirements of the industrial flow line.关键词
电表外观检定/工业自动化/深度学习/图像处理/场景文字识别Key words
electric meter appearance verification/industrial automation/deep learning/image processing/scene text recognition分类
信息技术与安全科学引用本文复制引用
安泰,赵森,谷依田,程大伟,张涛..基于OCR和点位配准的电表检定算法[J].电子器件,2025,48(2):339-345,7.
