一种复杂背景下的铁路货运车辆车号定位方法

Method for locating train number of railway freight vehicles in complex background

针对复杂背景下铁路货运车辆车号定位复杂、定位准确率低的问题,提出一种采用ResNet50作为基本特征提取网络,同时引入空残差块对学习样本进行多层特征融合,构造了一种新的特征提取网络和改进算法,提高目标检测网络的特征表达能力,实现了快速车号目标检测。实验数据集采用自建数据集,并通过三个对比实验验证了该方法的可靠性。当IoU阈值为0.5时,改进算法的平均精度值为97.1%,分别比F⁃VGG和F⁃ResNet50高9.4%和6.8%,同时采用改进算法对我国铁路常用不同车型货运车辆进行车号定位测试实验,从实验结果可以看出优化方法没有误分类或漏检。改进后的算法可以提高复杂背景下铁路货运车辆车号的定位精度,具有较强的泛化能力,对实现复杂背景下车辆车号快速定位具有一定的参考意义。

Railway freight vehicle number localization is a challenging problem and it suffers from an inferior recognition performance in complex background,so a new feature extraction network is established and an improved algorithm is proposed by using ResNet50 as the basic feature extraction network and by introducing null residual blocks to fuse the multi⁃layer features of the learning samples.This method aims to improve the feature expression ability of the object detection network and realize fast object detection of the vehicle number.The self⁃built data set is taken as the data set for the experiment.The effectiveness of the proposed method is verified by three contrastive experiments.When the threshold value of IOU(intersection over union)is 0.5,the average accuracy of the improved algorithm is 97.1%,which is higher than F⁃VGG and F⁃ResNet50 by 9.4%and 6.8%,respectively.The proposed method is also used to test the vehicle number location of different types of freight vehicles commonly used in China's railways.From the experimental results,it can be seen that the optimization method has no misclassification or missing detection cases.The improved algorithm can improve the localization accuracy of railway freight vehicle number in complex background.It has strong generalization ability.Therefore,it provides a positive example to realize rapid localization of vehicle number in complex background.