基于深度学习的液位仪表读数识别方法研究  

Liquid Level Instrument Recognition Based on Deep Learning

作　　者：李畅 王学军 Li Chang;Wang Xuejun(School of Information Science and Technology,Shijiazhuang Tiedao University,Shijiazhuang 050043,China)

机构地区：[1]石家庄铁道大学信息科学与技术学院,河北石家庄050043

出　　处：《石家庄铁道大学学报（自然科学版）》2023年第1期120-126,共7页Journal of Shijiazhuang Tiedao University(Natural Science Edition)

基　　金：河北省教育厅科学研究重点项目(ZD2016052);企业委托(50200020336)。

摘　　要：变电站对高铁安全运行至关重要,为准确获取高铁变电站液位仪表的准确读数,基于YOLOX-S提出KN-YOLOX-S深度学习网络模型。在骨干网络中,引入Ghost卷积代替传统卷积层,降低网络的参数量和计算量,实现模型的轻量化;在SPPBotteneck模块中增加KNSE注意力模块,提高网络对空间信息的敏感度,增强有效特征信息的提取能力。实验表明,KN-YOLOX-S比YOLOX-S模型在mAP@0.50上提高0.4%,mAP@0.5:0.95提高0.54%,同时推理速度提高近2倍,满足高铁变电站液位表实时检测要求。The substation is very important to the safe operation of high-speed railway.In order to accurately obtain the accurate reading of liquid level instrument in high-speed railway substation,this paper proposed KN-YOLOX-S deep learning network model based on YOLOX-S.In the backbone network,Ghost convolution was introduced to replace the traditional convolution layer,which reduced the network parameters and computation,and realized the lightweight of the model.The KNSE attention module was added to the SPPBotteneck module of the backbone network to improve the ability of the network,which extracted effective feature informationand enhanced the ability to extract effective feature information.The experiment showed that the KN-YOLOX-S model improved 0.4%compared with that of the YOLOX-S model in mAP@0.50,and mAP@0.5:0.95 improved 0.54%,and the reasoning speed almost tripled,meeting the real-time detection requirements of the liquid level gauge in the high-speed railway substation.

关 键 词：变电站 液位仪表 深度学习 实时检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]