摘要
研究基于inceptions结构神经网络的脱机手写汉字识别,提出了一种inception结构的改进方法,它具有结构更加简单、网络深度扩展更加容易、需要的训练参数量更少的优点。该方法在数据集CISIA-HWDB1.1上进行了实验验证,采用随机梯度下降优化算法,模型达到了96.95%的平均准确率。实验结果表明,使用改进的inception结构在图像分类上具有更好的鲁棒性,更容易扩展到其他应用领域。
This paper studied offline handwritten Chinese character recognition based on inception neural network.It proposed an improved inception structure,which took the advantages of simpler structure,easier network depth expansion and less training parameters.The method used the proposed structure to verify on dataset CISIA-HWDB1.1.The model achieved an average accuracy of 96.95% by using stochastic gradient descent optimization algorithm.Experimental result shows that the improved inception structure has better generalization performance and robustness in image classification,and can be easily extended to other applications.
作者
陈站
邱卫根
张立臣
Chen Zhan;Qiu Weigen;Zhang Lichen(School of Computers,Guangdong University of Technology,Guangzhou 510006,China)
出处
《计算机应用研究》
CSCD
北大核心
2020年第4期1244-1246,1251,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61873068)。
关键词
脱机手写汉字
卷积神经网络
INCEPTION
offline handwritten Chinese characters
convolutional neural network
inception
