燕山大学学报2024，Vol.48Issue(2)：165-170,6.DOI:10.3969/j.issn.1007-791X.2024.02.007
基于对抗学习的指针仪表自适应读数识别算法
Adaptive reading recognition algorithm of pointer instrument based on adversary learning
摘要
Abstract
To address the problems of high cost,inaccuracy,and low efficiency in manual reading of pointer instruments,a pose-invariant adaptive reading recognition algorithm of pointer instrument based on adversarial learning is proposed.This algorithm utilizes an object detection model to identify the position and attitude of the pointer instrument in the image,calibrates the pointer instrument and uses digital image processing technology for reading recognition.In order to improve the recognition effectiveness of the object detection model on pointer instrument images with different poses,a data augmentation technology based on adversarial learning is proposed,which constructs training data by optimizing rotation angles,translation distances,and scaled ratios of images that lead to inaccurate recognition,to improve the accuracy of the object detection model when the pointer instrument's pose changes.The research focuses on the SF6 pressure instrument commonly used inindustrial and mining enterprises,and experimental results show that the error of reading recognition is within 1%,which proves the effectiveness of the proposed algorithm.关键词
指针仪表/读数识别/目标检测/位姿不变/对抗学习Key words
pointer instrument/reading recognition/object detection/pose-invariant/adversarial learning分类
信息技术与安全科学引用本文复制引用
刘林,马云飞,王荷生,李宁..基于对抗学习的指针仪表自适应读数识别算法[J].燕山大学学报,2024,48(2):165-170,6.基金项目
河北省自然科学基金资助项目(E2020203027) （E2020203027）
国家电网河北省电力有限公司科技项目(kj2020-09) （kj2020-09）
