中文核心期刊
CSCD来源期刊
中国科技核心期刊
RCCSE中国核心学术期刊

重庆交通大学学报(自然科学版) ›› 2020, Vol. 39 ›› Issue (12): 1-5.DOI: 10.3969/j.issn.1674-0696.2020.12.01

• 交通+大数据人工智能 •    下一篇

基于深度学习的雾霾天气下交通标志识别

陈秀新,叶洋,于重重,张雪   

  1. (北京工商大学 食品安全大数据技术北京市重点实验室,北京 100048)
  • 收稿日期:2019-06-20 修回日期:2019-07-09 出版日期:2020-12-18 发布日期:2020-12-18
  • 作者简介:陈秀新(1979—),女,山东德州人,副教授,主要从事图像处理、视频信号处理方面的研究。E-mail:chenxx1979@126.com 通信作者:叶洋(1996—),男,安徽六安人,硕士研究生,主要从事图像处理方面的研究。E-mail:btbuyeyang@126.com
  • 基金资助:
    国家重点研发计划专项项目(2018YFC0807903)

Identification of Traffic Signs in Haze Weather Based on Deep Learning

CHEN Xiuxin, YE Yang, YU Chongchong, ZHANG Xue   

  1. (Beijing Key Laboratory of Big Data Technology for Food Safety, Beijing Technology and Business University, Beijing 100048, China)
  • Received:2019-06-20 Revised:2019-07-09 Online:2020-12-18 Published:2020-12-18
  • Supported by:
     

摘要: 针对雾霾天气下成像设备获取的图像质量较低导致交通标志难以识别这一现象,笔者提出了先去除雾霾后进行识别的办法。对雾霾图像首先通过深度学习算法IRCNN进行去雾霾处理,然后提出一种多通道卷积神经网络(Multi-channel CNN)模型对去雾霾后的图像进行识别。研究结果表明:IRCNN方法可有效去除雾霾,Multi-channel CNN模型识别效果好,设计的Multi-channel CNN模型的识别率在本次实验的数据集上达到100%,具有很好的泛化性和适应性。

 

关键词: 交通工程, 智能交通, 去雾霾, 交通标志识别, IRCNN , 多通道卷积神经网络

Abstract: In view of the phenomenon that the image quality acquired by imaging equipment was low in haze weather, which caused difficulty to identify traffic signs, the method of first removing haze and then identifying was proposed.Firstly,the haze image was processed by deep learning algorithm IRCNN,and then a multi-channel convolutional neural network (multi-channel CNN) model was proposed to recognize the image after haze removal.The research results show that IRCNN method can effectively remove haze,and the multi-channel CNN model has a good recognition effect. The recognition rate of the designed multi-channel CNN model reaches 100% on the data set of this experiment, which has good generalization and adaptability.

Key words: traffic engineering, intelligent transportation, haze removing, identification of traffic signs, IRCNN, multi-channel CNN

中图分类号: