摘要
针对车牌字符类别多、背景复杂等特点,以卷积神经网络LeNet-5模型为基础,通过去除全连接层F6层以及增加卷积层C1和C3层特征图的数目改进网络结构,通过字符分割,大小归一化、去除噪声、二值化、字符区居中、去除复杂背景等预处理,构建神经网络模型。结果表明:简化后的LeNet-5神经网络模型比传统的LeNet-5神经网络模型更为简单,其车牌字符识别算法准确率为99.96%。该研究对提高车牌字符识别的准确性提供了一定的参考。
This paper is a response to many types of license plates and more complex background. The study based on the LeNet-5 model of convolution neural network involves improving the network structure by removing F 6 layer of full connection layer and increasing the number of C 1 and C 3 layer feature maps of convolution layer;constructing the neural network model by preprocessing methods such as character segmentation, size normalization, noise removal, binarization, character area centralization, and complex background removal. The results show that the simplified LeNet-5 neural network model proves simpler than the traditional LeNet-5 neural network model, and could provide the license plate character recognition algorithm with the accuracy of 99.96%. This study could serve as a reference for improving the accuracy of license plate character recognition.
作者
赵艳芹
童朝娣
张恒
Zhao Yanqin;Tong Chaodi;Zhang Heng(School of Computer & Information Engineering, Heilongjiang University of Science &Technology, Harbin 150022, China)
出处
《黑龙江科技大学学报》
CAS
2019年第3期382-386,共5页
Journal of Heilongjiang University of Science And Technology
基金
中国煤炭工业协会指导项目(MTKJ2014-275)
