基于循环一致对抗学习的段码液晶仪表读数识别方法  

A reading recognition method for seven-segment LCD meters based on cycle-consistent adversarial learning

作　　者：惠毅 徐望明[1,2] 叶胜 HUI Yi;XU Wangming;YE Sheng(College of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China;Engineering Research Center for Metallurgical Automation and Detecting Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China)

机构地区：[1]武汉科技大学信息科学与工程学院,湖北武汉430081 [2]武汉科技大学冶金自动化与检测技术教育部工程研究中心,湖北武汉430081

出　　处：《武汉科技大学学报》2025年第1期59-69,共11页Journal of Wuhan University of Science and Technology

基　　金：国家自然科学基金项目(51805386);武汉科技大学冶金自动化与检测技术教育部工程研究中心开放课题(MADTOF2021B02).

摘　　要：为解决自然场景中仪表图像受复杂光照、背景及其他随机干扰所造成的读数识别难题,提出一种基于循环一致对抗学习的段码液晶仪表读数识别方法。该方法使用M-CRAFT作为字符检测网络,获取段码读数区域和段码字符位置,通过后处理算法分割得到段码字符子图像,然后以循环一致对抗域自适应网络CYCADA为基础,增加感知损失和重构损失并设计分类器,构建了一种改进的字符背景分离与识别网络I-CYCADA,用于将具有复杂背景的段码字符图像转换为清晰直观的二值图像,从而使读数识别变得更为简单和准确。为了验证该方法的有效性,构建了由多种复杂场景下的液晶段码仪表图像组成的数据集。实验结果表明,I-CYCADA可与不同CNN分类网络结合使用,均能提升转换后的段码字符图像的识别准确率,本文方法在自建数据集上的字符级识别准确率达95.39%,完整读数识别准确率达86.65%,有效改善了对困难样本的识别效果,且轻量级设计可满足实时性要求。To solve the difficult problem of reading recognition caused by complex illumination,background and other random interference to meter images in natural scenes,a novel method for seven-segment LCD meter readings based on cycle-consistent adversarial learning was presented.M-CRAFT was used as the character detecting network to obtain the reading region and the positions of the segmental code characters,and the sub-images of the characters were segmented via post-processing.Then,based on cycle-consistent adversarial domain adaption(CYCADA)network,an improved network for background separation and character recognition,named I-CYCADA,was constructed by considering the perceptual loss and the reconstruction loss as well as designing a classifier.I-CYCADA converted the segmental code character image with complex background into a clear and intuitive binary image,which made reading recognition simpler and more accurate.To verify the validity of the proposed method,a dataset composed of images of seven-segment LCD meters in a variety of complex scenarios was built.Experimental results show that I-CYCADA network can be used in combination with different CNN classification networks to improve the recognition accuracy of the converted images of segmental code characters.The proposed method achieves a character-level recognition accuracy of 95.39%and a whole reading recognition accuracy of 86.65%on the self-constructed dataset.The recognition effect of difficult samples is effectively improved,and the lightweight design can meet the real-time requirements.

关 键 词：液晶仪表 读数识别 对抗学习 域自适应 感知损失 重构损失 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术]