基于残差网络的数字仪表检测方法研究  

Research on Digital Instrument Detection Method Based on Residual Network

作　　者：隋涛[1] 冯永新[1] 史帛加 钱博[1] SUI Tao;FENG Yong-xin;SHI Bo-jia;QIAN Bo(Shenyang University of Technology,Shenyang Liaoning 110158,China)

机构地区：[1]沈阳理工大学,辽宁沈阳110158

出　　处：《计算机仿真》2024年第11期347-352,共6页Computer Simulation

基　　金：国家自然基金项目(61971291);辽宁省教育厅科学研究项目(LJKZ0242)。

摘　　要：数字式仪表在工业应用和实验室测试中都有着非常广泛的应用,文本检测对数字式仪表读数识别的准确率起到重要作用,所以优化读数区域的文本候选框的检测性能有着重要意义。为了获取不同型号和不同种类数字仪表图像的读数文本区域,基于残差网络设计带有不同卷积核数量的标准残差结构。然后将应用该结构的RegNet模型集合和DBNet模型相结合,搭建了文本检测模型集合,具有快速、高效并且可以适用于大多数型号的数字仪表等优势。经实验表明,改进的检测模型在多个性能指标上均优于传统模型,说明该网络可以准确的检测出数字仪表读数的文本区域。Digital instrument is widely used in industrial applications and laboratory tests.Text detection plays an important role in the accuracy of digital instrument reading recognition,so it is of great significance to optimize the de-tection performance of text candidate box in the reading area.In order to obtain the reading text areas of different models and different kinds of digital instrument images,the standard residual structure with different number of convo-lution cores is designed based on the residual network.Then,the RegNet model set and DBNet model are combined to build the text detection model set,which has the advantages of fast,efficient and can be applied to most models of dig-ital instruments.The experiment shows that the improved detection model is better than the traditional model in many performance indicators,indicating that the network can accurately detect the text area of the digital instrument reading.

关 键 词：深度学习 仪表检测 自适应阈值 文本检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]