铁路车号CMOS图像传感信息智能识别方法  

Intelligent Recognition Method of Railway Vehicle Number CMOS Image Sensing Information

作　　者：刘为民 LIU Weimin(Harbin Freight Car Depot,China Railway Harbin Group Co.,Ltd.,Harbin 150080,China)

机构地区：[1]中国铁路哈尔滨局集团有限公司哈尔滨车辆段,黑龙江哈尔滨150080

出　　处：《机械制造与自动化》2024年第3期275-279,共5页Machine Building & Automation

摘　　要：CMOS传感器在采集铁路车号图像过程中存在多种外界因素的混合干扰,图像重复区域增多,增加了铁路车号识别复杂度。为此提出基于CMOS传感器的铁路车号识别方法。利用CMOS传感器采集多种因素干扰条件下的铁路车号图像,利用深度卷积神经网络对铁路车号图像重复区域进行检测,剔除无用矩形框,确定有效的车号区域。利用维纳滤波器对铁路车号预选框区域去噪处理,有效避免干扰因素对于铁路车号图像质量的不利影响。搭建铁路车号识别模型,利用多任务学习方法对该模型进行训练,将去噪处理后的图像输入该模型中,得到铁路车号识别结果。实验结果表明:所提方法的铁路车号识别精度高和时间短,实际应用效果好。Rearding the increases of image repetition area and complexity of railway vehicle number recognition caused by mixed interferences of various external factors in the process of collecting railway vehicle number image with CMOS sensor,a railway vehicle number recognition method based on CMOS sensor is proposed.The CMOS sensor is used to collect the railway vehicle number image under the interference of various factors,and the deep convolution neural network is applied to detect the duplicate area of the railway vehicle number image,eliminate the useless rectangular frame,and determine the effective vehicle number area.With the Wiener filter,the railway vehicle number pre-selection frame area is de-noised to effectively avoid the adverse impact of interference factors on the railway vehicle number image quality.A railway vehicle number recognition model is built,trained by multi-task learning method,input the de-noised image to get the railway vehicle number recognition results.The experimental results show that the train number recognition by the proposed method has high accuracy and short time with good practical application effect.

关 键 词：CMOS传感器 铁路车号 智能识别 维纳滤波器 多任务学习 

分 类 号：U271[机械工程—车辆工程] TP391.41[交通运输工程—载运工具运用工程]