现代电子技术2025，Vol.48Issue(5)：169-174,6.DOI:10.16652/j.issn.1004-373x.2025.05.026
基于改进U2-Net的指针式仪表读数识别方法
Pointer meter reading recognition method based on improved U2-Net
摘要
Abstract
In view of the uneven illumination and the fact that the geometric distortion affects the extraction of key areas and the accuracy rate of reading recognition for instrument dials,a method based on improved U2-Net for instrument correction and reading recognition is proposed.Firstly,a U2-Net model with an integrated coordinate attention mechanism is employed to identify and segment the instrument dial area.Subsequently,an improved perspective transformation technique is used for geometric correction of the dial area,achieving geometric corrections for meters of various shapes.An adaptive MSRCR algorithm is then introduced to correct the unevenly illuminated areas of the dial.Finally,the U2-Net is used to segment the corrected image for pointers and key scale points.The pointer line is fitted with PCA to obtain the pointer deflection angle,and the accurate reading is calculated with the angular method.Experimental results show that the proposed method can effectively recover key information areas of meters under conditions of insufficient or excessive illumination,and can correct inclined meters of different shapes effectively.In addition,it enhances the accuracy of pointer and scale information extraction,with a reading error rate lower than 0.89% .关键词
指针式仪表/光照校正/失真校正/U2-Net/仪表读数/MSRCRKey words
pointer meter/illumination correction/distortion correction/U2-Net/meter reading/MSRCR分类
信息技术与安全科学引用本文复制引用
李丽,乔逸天,黄小龙,谢维成,蒋文波..基于改进U2-Net的指针式仪表读数识别方法[J].现代电子技术,2025,48(5):169-174,6.基金项目
四川省科技成果转移转化项目(2022ZHCG0099) （2022ZHCG0099）
国家自然科学基金项目(61875166) （61875166）
