工程科学学报.第42卷，第11期：1525-1533.2020年11月 Chinese Journal of Engineering,Vol.42,No.11:1525-1533,November 2020 https://doi.org/10.13374/j.issn2095-9389.2019.12.05.001;http://cje.ustb.edu.cn 基于深度学习的高效火车号识别 王志明区，刘志辉”，黄洋科)，邢宇翔 1)北京科技大学计算机与通信工程学院.北京1000832)清华大学工程物理系，北京100084 ☒通信作者，E-mail:wangzhiming@ustb.edu.cn 摘要基于高性能的YOLO3目标检测算法，提出一种分阶段高效火车号识别算法.整个识别过程分为两个阶段：第一阶 段在低分辨率全局图像中检测出火车号区域位置：第二阶段在局部高分辨率图像中检测出组成火车号的字符，根据字符的空 间位置关系搜索得到12位火车号，并利用每个字符的识别置信度及火车号编码规则进行校验得到最终火车号.另外，本文提 出一种结合批一化因子和滤波器相关度的剪枝算法，通过对两个阶段检测模型的剪枝，在保证识别准确率不降（实验中略有 提升)的条件下降低了存储空间占用率和计算复杂度.在现场采集的1072幅火车号图像上的实验结果表明，本文提出的火车 号识别算法达到了96.92%的整车号识别正确率，平均识别时间仅为191ms 关键词模式识别：火车号识别：深度学习：神经网络：目标检测：模型剪枝 分类号TP391 Efficient wagon number recognition based on deep learning WANG Zhi-ming,LIU Zhi-hui,HUANG Yang-ke,XING Yu-xiang? 1)School of Computer and Communication Engineering,University of Science and Technology Beijing,Beijing 100083,China 2)Department of Engineering Physics,Tsinghua University,Beijing 100084,China Corresponding author,E-mail:wangzhiming @ustb.edu.cn ABSTRACT The automatic recognition of a wagon number plays an important role in railroad transportation systems.However,the wagon number character only occupies a very small area of the entire wagon image,and it is often accompanied by uneven illumination, a complex background,image contamination,and character stroke breakage,which makes the high-precision automatic recognition difficult.In recent years,object detection algorithm based on deep learning has made great progress,and it provides a solid technical basis for us to improve the performance of the train number recognition algorithm.This paper proposes a two-phase efficient wagon number recognition algorithm based on the high-performance YOLOv3 object detection algorithm.The entire recognition process is divided into two phases.In the first phase,the region of the wagon number in an image is detected from a low-resolution global image; in the second stage,the characters are detected in a high-resolution local image,formed into the wagon number according to their spatial position,and the final wagon number is obtained after verification based on the recognition confidence of each character and international wagon number coding rules.In addition,we proposed a new deep learning network-pruning algorithm based on the batch normalize scale factor and filter correlation.The importance of every filter was computed by considering the correlation between filter weights and the scale factor generated via batch normalization.By pruning and retraining the region detection model and character detection model,the storage space occupation and computational complexity were reduced without sacrificing recognition accuracy (which is even slightly improved in our experiment).Finally,we tested the proposed two-phase wagon number recognition algorithm on 1072 images from practical engineering application scenarios,and the results show that the proposed algorithm achieves96.9%of the overall correct ratio (here,"correct"means all 12 characters are detected and recognized correctly),and the average recognition time is only 191 ms. 收稿日期：2019-12-05基于深度学习的高效火车号识别 王志明1) 苣，刘志辉1)，黄洋科1)，邢宇翔2) 1) 北京科技大学计算机与通信工程学院，北京 100083    2) 清华大学工程物理系，北京 100084 苣通信作者，E-mail：wangzhiming@ustb.edu.cn 摘    要    基于高性能的 YOLOv3 目标检测算法，提出一种分阶段高效火车号识别算法. 整个识别过程分为两个阶段：第一阶 段在低分辨率全局图像中检测出火车号区域位置；第二阶段在局部高分辨率图像中检测出组成火车号的字符，根据字符的空 间位置关系搜索得到 12 位火车号，并利用每个字符的识别置信度及火车号编码规则进行校验得到最终火车号. 另外，本文提 出一种结合批一化因子和滤波器相关度的剪枝算法，通过对两个阶段检测模型的剪枝，在保证识别准确率不降（实验中略有 提升）的条件下降低了存储空间占用率和计算复杂度. 在现场采集的 1072 幅火车号图像上的实验结果表明，本文提出的火车 号识别算法达到了 96.92% 的整车号识别正确率，平均识别时间仅为 191 ms. 关键词    模式识别；火车号识别；深度学习；神经网络；目标检测；模型剪枝 分类号    TP391 Efficient wagon number recognition based on deep learning WANG Zhi-ming1) 苣 ，LIU Zhi-hui1) ，HUANG Yang-ke1) ，XING Yu-xiang2) 1) School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China 2) Department of Engineering Physics, Tsinghua University, Beijing 100084, China 苣 Corresponding author, E-mail: wangzhiming@ustb.edu.cn ABSTRACT    The automatic recognition of a wagon number plays an important role in railroad transportation systems. However, the wagon number character only occupies a very small area of the entire wagon image, and it is often accompanied by uneven illumination, a  complex  background,  image  contamination,  and  character  stroke  breakage,  which  makes  the  high-precision  automatic  recognition difficult. In recent years, object detection algorithm based on deep learning has made great progress, and it provides a solid technical basis for us to improve the performance of the train number recognition algorithm. This paper proposes a two-phase efficient wagon number  recognition  algorithm  based  on  the  high-performance  YOLOv3  object  detection  algorithm.  The  entire  recognition  process  is divided into two phases. In the first phase, the region of the wagon number in an image is detected from a low-resolution global image; in the second stage, the characters are detected in a high-resolution local image, formed into the wagon number according to their spatial position,  and  the  final  wagon  number  is  obtained  after  verification  based  on  the  recognition  confidence  of  each  character  and international wagon number coding rules. In addition, we proposed a new deep learning network-pruning algorithm based on the batch normalize scale factor and filter correlation. The importance of every filter was computed by considering the correlation between filter weights  and  the  scale  factor  generated via batch  normalization.  By  pruning  and  retraining  the  region  detection  model  and  character detection  model,  the  storage  space  occupation  and  computational  complexity  were  reduced  without  sacrificing  recognition  accuracy (which is even slightly improved in our experiment). Finally, we tested the proposed two-phase wagon number recognition algorithm on 1072 images from practical engineering application scenarios, and the results show that the proposed algorithm achieves 96.9% of the overall correct ratio (here, “correct” means all 12 characters are detected and recognized correctly), and the average recognition time is only 191 ms. 收稿日期: 2019−12−05 工程科学学报，第 42 卷，第 11 期：1525−1533，2020 年 11 月 Chinese Journal of Engineering, Vol. 42, No. 11: 1525−1533, November 2020 https://doi.org/10.13374/j.issn2095-9389.2019.12.05.001; http://cje.ustb.edu.cn