摘要
为克服传统手写汉字识别采用人工提取特征的局限,借鉴GoogLeNet网络搭建了一个卷积神经网络.针对训练样本不足的问题,采用随机弹性变换算法扩充了训练数据.结果表明:新的网络结构与随机弹性变换算法配合使用,与采用传统仿射变化扩充样本的模型比较,极大提升了识别的正确率,并具有较强的泛化能力.
In order to overcome the limitation of manual feature extraction in traditional handwritten Chinese character recognition,a convolution neural network suitable for handwritten Chinese character recognition is proposed in this paper according to GoogLeNet.To solve the problem of insufficient training samples,the random elastic transform algorithm is used to expand the training data sets.The result shows that the new network structure is used in conjunction with the stochastic elastic transform algorithm,which greatly improves the recognition accuracy and has strong generalization ability.
作者
林恒青
郑晓斌
王麟珠
戴立庆
LIN Heng-qing;ZHENG Xiao-bing;WANG Lin-zhu;DAI Li-qing(Department of Mechanical Engineering,Fujian Chuanzheng Communications College,Fuzhou 350007,China)
出处
《兰州工业学院学报》
2020年第3期62-67,共6页
Journal of Lanzhou Institute of Technology
基金
2017年福建省教育厅科技项目(JAT170942)。
关键词
手写汉字识别
深度学习
卷积神经网络
弹性变换
handwritten Chinese character recognition
deep learning
convolution neural networks
elastic transform
