【作者】 郭哲； 孙博伟； 谢力仁； 王毅； 樊养余；

【Author】 GUO Zhe;SUN Bowei;XIE Liren;WANG Yi;FAN Yangyu;School of Electronics and Information,NorthwesternPolytechnical University;

【机构】 西北工业大学电子信息学院；

【摘要】 手写字体识别是机器学习、人工智能等领域的重要课题,基于深度学习的手写字体识别是近年来的研究热点。目前已有的用于手写字体识别的深度学习网络框架主要有CNN网络和LeNet-5网络。针对现有网络对手写连体字符识别率不高的问题,本文提出了基于优化策略的改进方法,对LeNet-5网络通过选用合适的优化器,设置损失函数,使用正则化以及Dropout等方法改善网络过拟合的问题,并进一步提高手写字体识别的准确率。基于MNIST数据集的识别实验结果表明,本文提出的改进的基于深度学习的手写字体识别方法,在测试集的识别准确率达到99.3%,有效提高了原方法的识别准确率。更多还原

【Abstract】 Handwritten character recognition is an important topic in machine learning,artificial intelligence and other fields.Handwritten character recognition based on deep learning is a research hotspot in recent years.At present,the existing deep learning network frameworks for handwriting character recognition mainly include CNN network and LeNet-5 network.To solve the problem that the recognition rate of handwriting concatenation characters in the existing network is not high,an improved method based on optimization strategy is proposed in this paper.For LeNet-5 network,appropriate optimizer is selected,loss function is set,regularization and Dropout are used to improve the problem of network overfitting,as a result,the accuracy of handwritten character recognition is further improved.The experimental results based on MNIST dataset show that the improved handwritten character recognition method based on deep learning proposed in this paper has a recognition accuracy of 99.3%,which effectively improves the recognition accuracy of the original method.更多还原