基于深度学习的车牌识别系统研究  

Research on License Plate Recognition System Based on Deep Learning

作　　者：孙小广 万若楠 Sun Xiaoguang;Wan Ruonan(College of Electronic Information Engineering,Guangzhou City University of Technology,Guangzhou 510800,China)

机构地区：[1]广州城市理工学院电子信息工程学院,广州510800

出　　处：《科技通报》2024年第6期32-35,共4页Bulletin of Science and Technology

摘　　要：本文针对目前车牌识别受限于光照、角度、场地等环境影响,开展基于深度学习的车牌识别研究。车牌定位前先对车牌图像进行灰度化处理、均值滤波消除噪声、边缘检测、二值化等预处理,结合几何、颜色等多项特征进行定位;再使用垂直投影法,找到每个字符的边界区域,逐一进行字符切割;最后搭建卷积神经网络,构建训练集和测试集,通过深度学习,识别和输出车牌字符,并在MATLAB上测试和仿真,准确率达到98.6%。To address the current limitations of license plate recognition due to environmental factors such as lighting,angle,and location,research on license plate recognition based on deep learning is constantly deepening.Before li-cense plate localization,the license plate image was preprocessed with grayscale processing,eliminating noise by mean filtering,edge detection,binarization,and multiple features such as geometry and color were combined for lo-calization;Then vertical projection method was used to find the boundary area of each character and perform character segmentation one by one;Finally,a convolutional neural network was created to construct a training and testing set.Through deep learning,license plate characters were recognized and output,and tested and validated on MATLAB with an accuracy of 98.6%.

关 键 词：车牌定位 车牌识别 卷积神经网络 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]