DBN和CNN融合的脱机手写汉字识别  

Offline Handwritten Chinese Character Recognition Based on DBN and CNN Fusion Model

作　　者：李兰英[1] 周志刚 陈德运[1] LI Lan-ying;ZHOU Zhi-gang;CHEN De-yun(School of Computer Science and Technology, Harbin University of Science and Technology, Harbin 150080, China)

机构地区：[1]哈尔滨理工大学计算机科学与技术学院,哈尔滨150080

出　　处：《哈尔滨理工大学学报》2020年第3期137-143,共7页Journal of Harbin University of Science and Technology

基　　金：国家自然科学基金青年基金(61501147)。

摘　　要：针对脱机手写汉字形近字多,提取特征难,识别不准的问题,提出了一种卷积神经网络和深度信念网络的融合模型。首先在数据集上分别训练卷积神经网络和深度信念网络,发现二者的综合TOP-2准确率可达到99.33%。利用卷积神经网络和深度信念网络在图像分析中各自的优势,采用了一种融合比较策略,在两者的TOP-2分类中尽可能准确地取出一个分类结果以提高识别的能力。实验结果表明:卷积神经网络和深度信念网络的融合模型比单独使用卷积神经网络和深度信念网络具有更好的识别效果。Aiming at the problem that some offline handwritten Chinese characters are similar in shape and it is difficult to extract the feature of characters and the recognition is not accurate,a convolutional neural network and deep belief network fusion model is proposed.Firstly,the convolutional neural network and the deep belief network are trained on the dataset respectively.It is found that the comprehensive TOP-2 accuracy of the both can reach 99.33%.Using the advantages of convolutional neural networks and deep belief networks in image analysis,a fusion comparison strategy is adopted to extract a classification result as accurately as possible in the TOP-2 classification of the two to improve the recognition ability.The experimental results show that the fusion model of convolutional neural network and deep belief network has better recognition effect than convolutional neural network and deep belief network.

关 键 词：卷积神经网络 深度信念网络 脱机手写汉字 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]