铁路货车号码图像精准识别方法研究  

Research on Accurate Recognition Method of Railway Freight Car Number Image

作　　者：吴志伟 高达 刘慧贤 张佳欢 WU Zhiwei;GAO Da;LIU Huixian;ZHANG Jiahuan(Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Center of National Railway Intelligent Transportation System Engineering and Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司电子计算技术研究所,北京100081 [2]中国铁道科学研究院集团有限公司国家铁路智能运输系统工程技术研究中心,北京100081

出　　处：《铁道货运》2023年第11期61-68,共8页Railway Freight Transport

基　　金：中国国家铁路集团有限公司科技研究开发计划课题(J2021X005)。

摘　　要：我国铁路货运持续增长,但货场车号员需要人工核对车号,使其作业中存在效率低、劳动强度大等问题。智能识别铁路货车号码是提升整体运行效率的重要措施,目前货运站场的号码高精度识别受到多种干扰因素的影响,故提出一种基于深度学习的多类别号码精准识别方法,以解决铁路平车、棚车、罐车号码识别问题。首先,采用颜色特征的图像预处理方法,强化号码区域对比度,减弱磨损影响;其次,以Faster R-CNN为基本框架,通过减少网络参数量、多尺度训练和空间姿态纠正网络,提升训练效率和检测准确率;最后,使用CRNN网络进行车号识别。实验结果表明,该方法能有效提高铁路货车车号检测和识别的速度及准确度。Rail freight traffic in China has been growing steadily.However,the manual verification and reporting of car numbers by yard staff is inefficient and labor-intensive.Intelligent detection of rail freight car numbers is critical to improving overall operational efficiency.However,high-precision recognition of car numbers in freight yards is affected by various interfering factors.To address this problem,this paper proposed a deep learning-based method for the accurate recognition of multiple categories of car numbers,including flatbed,car,and oil tank car numbers,in railway freight yards.Firstly,a color feature-based image pre-processing method was adopted to enhance the contrast of the number area and reduce the effect of wear.Secondly,the paper employed the Faster R-CNN framework to reduce network parameters,enhance target area features,use multi-scale training strategies and a spatial pose correction network to improve training efficiency and recognition accuracy.Finally,this paper used a CRNN network for accurate recognition of the detected number images.Experimental results show that this method can effectively improve the speed and accuracy of railway freight car number detection and recognition.

关 键 词：号码识别 铁路货运 Faster R-CNN 深度学习 图像预处理 

分 类 号：U294[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]