基于深度学习的轻量化车牌检测算法研究
Research on Deep Learning Based Lightweight License Plate Detection Algorithm
杜铭辉 1吴林煌 1苏喆1
作者信息
- 1. 福州大学 物理与信息工程学院,福建 福州 350108
- 折叠
摘要
针对现有车牌检测算法存在的模型参数量过大、实时性差和检测效果不佳等问题,提出一种基于深度学习的轻量化车牌检测网络(Lightweight License Plate Detection Networks,LW-LPDNet)模型.该模型以PP-LCNet作为骨干网络,大幅减少模型参数量,同时融入压缩-激励网络(Squeeze and Excitation Networks,SE-Net)注意力模块,增加车牌信息的通道权重.最后,引入SimSPPF和GSConv,对多尺度特征进行融合,增大感受野,进一步提高检测准确率.通过对模型进行训练和测试,LW-LPDNet在中国城市停车数据集(Chinese City Parking Dataset,CCPD)上获得98.9%的平均精确率,优于其他车牌检测方法,且模型参数量仅有0.13 MB,检测速度达到243 f·s-1,具备较高的实时性.
Abstract
Aiming at the problems of the existing License Plate Detection algorithms,such as large number of model parameters,poor real-time performance and poor detection effect,a Lightweight License Plate Detection Networks(LW-LPDNet)model based on deep learning was proposed.In this model,PP-LCNet is used as the backbone network,the number of model parameters is greatly reduced,and the attention module of Squeeze and Excitation Networks(SE-Net)is integrated to increase the channel weight of license plate information.Finally,SimSPPF and GSConv were introduced to fuse the multi-scale features,enlarge the receptive field,and further improve the detection accuracy.Through the training and testing of the model,LW-LPDNet obtained an average accuracy rate of 98.9%on the Chinese City Parking Dataset(CCPD),which was better than other license plate detection methods,and the model parameter size was only 0.13MB.The detection speed reaches 243 f·s-1,and the real-time performance is high.
关键词
深度学习/轻量化/注意力/车牌检测Key words
deep learning/lightweight/attention/license plate detection引用本文复制引用
出版年
2024
NETL
NSTL
万方数据