摘要
针对车牌字符图像进行特征提取与识别算法的研究。使用BP神经网络识别算法建立车牌字符图像的识别。由于常规BP神经网络算法本身存在训练过程容易陷入局部最小值、收敛效率低以及网络结构参数不易确定等缺点。另外,BP网络结构参数的选取对算法的性能影响很大,而参数的选取通常是根据经验公式选取的,存在很大的随机性和盲目性,使算法的性能无法得到保证。因此该文使用收敛速度快、适用于全局搜索的PSO优化算法对BP神经网络算法的性能进行优化,研究一种双粒子群优化的改进BP神经网络算法。最后通过车牌识别实验对识别算法进行研究,结果表明,通过对神经网络算法进行改进,使用其建立汉字识别模型、字母识别模型以及混合识别模型的识别准确率均优于常规神经网络算法建立的模型,具有较好的识别性能。
The feature extraction and recognition algorithm of license plate character image are studied. The BP neural net?work recognition algorithm is used to identify the license plate character image. Since the conventional BP neural network algo?rithm is easy to fall into local minimum value and low convergence efficiency,and difficult to determine the network structureparameters in training process,moreover,since the selection of BP network structure parameters has great influence on the per?formance of the algorithm,and the parameters selection is usually based on the empirical formula and has prodigious random?ness and blindness,the performance of the algorithm can’t be guaranteed. Therefore,the PSO algorithm is used to optimize theperformance of BP neural network algorithm,which has fast convergence rate and is suitable for global search. The improved BPneural network algorithm for the dual particle swarm optimization is studied. The recognition algorithm is studied with the experi?ment of license plate recognition. The results show that the recognition accuracy of Chinese characters recognition model,lettersrecognition model and mixed recognition model established by means of the improved neural network algorithm is superior to thatof the model established with conventional neural network algorithm,and the recognition algorithm has good recognition perfor?mance.
作者
李战明
杨红红
LI Zhanming;YANG Honghong(College of Electrical and Information Engineering,Lanzhou University of Technology,Lanzhou 730050,China;Lanzhou University of Technology,Lanzhou 730050,China)
出处
《现代电子技术》
北大核心
2016年第16期102-104,107,共4页
Modern Electronics Technique
基金
教育部博士点基金(20106201110003)
关键词
车牌字符识别
特征提取
神经网络
粒子群优化算法
license plate character recognition
feature extraction
neural network
particle swarm optimization algorithm