张晓丽,董 昱.面向铁路货车车号定位的 Faster R-CNN卷积神经网络[J].电子测量与仪器学报,2020,34(10):192-200 |
面向铁路货车车号定位的 Faster R-CNN卷积神经网络 |
Faster R-CNN convolutional neural network for the location of freight train number |
|
DOI: |
中文关键词: 货车车号定位 Faster R-CNN 卷积神经网络 特征增强 |
英文关键词:train number location Faster R-CNN convolutional neural network feature enhancement |
基金项目:国家自然科学基金(61763023)资助项目 |
|
|
摘要点击次数: 501 |
全文下载次数: 636 |
中文摘要: |
为了解决传统算法对于铁路货运列车车号识别准确率不高问题,提出了一种面向铁路货车车号定位的 Faster R-CNN 神
经网络。 通过调整特征提取网络的相关尺寸参数及连接方式增强了最后一层卷积特征图的细节特征。 并采用 k-means++聚类
算法求取车号区域长宽比改进 anchor 尺寸设计,使目标检测框与实际车号区域更加贴合。 实验过程中,采用了数据增广、
dropout 方法提升网络的鲁棒性。 结果显示,改进 Faster R-CNN 网络在铁路货车车号定位精度达到了 93. 15%,召回率 90. 76%,
综合 F1 指标 91. 94%,也说明该方法能够对铁路货车车号准确定位,并为车号识别过程提供可靠的数据支持。 |
英文摘要: |
In order to solve the problem of low accuracy of traditional algorithm for train number identification of railway freight trains,
Faster R-CNN neural network for train number location of railway freight trains is proposed. The detailed features of the final convolution
feature map are enhanced by adjusting the relevant size parameters and connection mode of the feature extraction network. The k-means
++ clustering algorithm is used to calculate the length width ratio of the train number area. The improved anchor size design makes the
target detection frame more suitable for the actual train number area. In the experiment, data augmentation and dropout are used to
improve the robustness of the network. The results show that the improved Faster R-CNN network has achieved 93. 15% accuracy in the
location of railway freight train number, 90. 76% recall rate and 91. 94% comprehensive F1 index. It also shows that this method can
accurately locate the railway freight train number and provide reliable data support for the identification process. |
查看全文 查看/发表评论 下载PDF阅读器 |