复杂字符干扰场景下铁路集装箱箱号快速定位方法研究  

Fast location method of railway container number in complex character interference scenario

作　　者：张添添 周书民[1] 蓝贤桂[1] ZHANG Tiantian;ZHOU Shumin;LAN Xiangui(Jiangxi Engineering Research Center for New Energy Technique and Equipment,East China University of Technology,Nanchang 330013,China)

机构地区：[1]东华理工大学江西省新能源工艺及装备工程技术研究中心,江西南昌330013

出　　处：《现代电子技术》2023年第11期81-87,共7页Modern Electronics Technique

基　　金：国家自然科学基金项目(12165001)。

摘　　要：针对铁路集装箱箱号快速定位由于存在复杂的字符干扰,采用图像识别方法存在定位速度慢、精度低的问题,文中提出一种基于改进YOLOv3的集装箱箱号定位算法,该算法将主干网络替换为EfficientNetv2轻型网络,并根据数据特点改进了损失函数,利用规整通道剪枝实现了模型剪枝,增加了SPPF模块。实验结果表明:基于改进的YOLOv3算法模型大小仅有18.6 MB,相比YOLOv3模型而言减小了92%;定位准确率为97.4%,定位精度较YOLOv3提升了3.1%,同时能达到21.3 ms的定位速度。相较于YOLOv3和YOLOv3⁃Tiny模型,该模型更加适用于铁路集装箱箱号的快速智能识别。The rapid positioning of railway container number has the complex character interference,since the available image recognition method has the problems of slow positioning speed and low accuracy,a container number location algorithm based on improved YOLOv3 is proposed.With the algorithm,the main network is replaced with the light network EfficientNetv2,and the loss function is improved according to the data characteristics.The structured channel pruning is used to realize the model pruning,and the SPFF module is added.The experimental results show that the size of the model based on the improved YOLOv3 algorithm is only 18.6 MB,which is decreased by 92%in comparison with the YOLOv3 model;its positioning accuracy is 97.4%,which is increased by 3.1%in comparison with the YOLOv3;its positioning speed can reach 21.3 ms.In comparison with the models of YOLOv3 and YOLOv3⁃Tiny,this model is more suitable for fast and intelligent identification of container number in railway container transportation industry.

关 键 词：集装箱箱号 定位算法 EfficientNetv2 模型剪枝 YOLOv3 SPPF模块 图像预处理 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]