基于深度学习的数显仪表字符识别  

Character Recognition of Digital Display Instrument Based on Deep Learning

作　　者：朱立倩 ZHU Li-qian(School of Computer Science and Technology,China University of Petroleum,Qingdao 266580,China)

机构地区：[1]中国石油大学(华东)计算机科学与技术学院,山东青岛266580

出　　处：《计算机技术与发展》2020年第6期141-144,共4页Computer Technology and Development

基　　金：国家自然科学基金(61402533);山东省自然科学基金资助项目(ZR2019MF049)。

摘　　要：在许多工业场景中,需要记录仪表的数据,将数据录入到电脑,这不仅耗时耗力,而且两次的转录可能导致错误的发生。为了提高监控效率,需要对数显仪表数据进行自动识别。针对传统字符分割方法适应性差,准确度低的不足,提出了一种基于深度学习的自动识别数显仪表字符的方法,由字符区域定位网络及字符识别网络构成。字符区域定位网络为改进的Faster R-CNN,将Faster R-CNN的骨干网络改为ResNeXt-101,感兴趣区域池化操作改为精确的感兴趣区域池化操作,以提高分类及定位的准确性。字符识别网络由卷积神经网络和加入注意力机制的长短时记忆网络构成,注意力机制的加入提高了字符识别的准确性。以变压器直流电阻测试仪为具体应用对象,实验结果显示,该方法可以达到95%的准确率。In many industrial scenarios,it is necessary to record the data of the digital display instrument and then input it into the computer,which is not only time-consuming and laborious,but also two transcriptions may lead to errors.In order to improve the monitoring efficiency,it is necessary to recognize the data of the digital display instrument automatically.Aiming at the shortcomings of poor adaptability and low accuracy of traditional character segmentation methods,we propose a method of automatic recognition of digital display instrument characters based on deep learning,which is composed of character region location network and character recognition network.The character region location network is improved Faster R-CNN.The backbone network of Faster R-CNN is changed to ResNeXt-101,and the ROI-pooling is changed to precise ROI-pooling to improve the accuracy of classification and location.Character recognition network is composed of convolutional neural network and long short-term memory network with attention mechanism,which improves the accuracy of character recognition.Taking transformer DC resistance tester as a specific application object,the experimental results show that the proposed method can achieve 95%accuracy.

关 键 词：数显仪表 卷积神经网络 注意力机制 字符检测 字符识别 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]