摘要
车牌识别易受到天气或光照条件等不可预见干扰的影响。因此,文章提出了一种无需分割的车牌识别框架,该网络模型结合了先进的深度学习方法和设计思想,采用了深度可分离卷积来极大地降低计算量。与之前相比,它以更低的计算要求实现了更高的识别精度。在两个不同的数据集上对所提方法的有效性进行了评估,并获得了超过99%识别准确率和70以上的帧率,该方法稳健高效,值得推广。
In the real world,license plate recognition still faced many challenges and was affected by unforeseeable interference such as weather or lighting conditions.To this end,a segmentation-free license plate recognition framework was proposed,which combined advanced deep learning methods and design ideas,and adopted depth-wise separable convolutions to greatly reduce computational complexity.Compared with previous work,it achieved higher recognition accuracy with lower computational requirements.The effectiveness of the proposed method was evaluated on two different datasets,and over 99%recognition accuracy and a frame rate of over 70 were obtained,indicating that the method was not only robust but also efficient.
作者
林璐颖
LIN Luying(Electronic Information Department,Zhangzhou Institute of Technology,Zhangzhou,Fujian 363000,China)
出处
《九江学院学报（自然科学版）》
CAS
2024年第2期82-86,共5页
Journal of Jiujiang University：Natural Science Edition
基金
福建省教育科学“十四五”规划2022年度课题(编号FJJKGZ22-057)
福建省中青年教师教育科研项目(科技类)(编号JAT201269)
2024年漳州职业技术学院科研项目(编号zzyzj24005)的成果之一
关键词
车牌识别
深度学习
深度可分离卷积
license plate recognition
deep learning
depth-wise separable convolution
作者简介
林璐颖(1981-),福建漳州人,硕士,讲师,研究方向:计算机视觉和机器学习。Email:lly2022a@163.com。
