详细信息
文献类型:期刊文献
中文题名:基于CUDA和深度置信网络的手写字符识别
英文题名:Handwritten Character Recognition Based on CUDA and Deep Belief Networks
作者:陆军建[1];林家骏[1]
机构:[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237
年份:2015
卷号:41
期号:2
起止页码:210
中文期刊名:华东理工大学学报(自然科学版)
外文期刊名:Journal of East China University of Science and Technology
收录:CSTPCD;;Scopus;北大核心:【北大核心2014】;CSCD:【CSCD2015_2016】;
语种:中文
中文关键词:GPU;CUDA;手写字符识别;深度置信网络
外文关键词:GPU; CUDA; handwritten character recognition; Deep Belief Networks
摘要:为了应对海量的字符(手写)识别,提出了一种将统一计算设备架构(Compute Unified Device Architecture,CUDA)和深度置信网络相结合的方法进行手写字符识别。该方法结合受限玻尔兹曼机和反向传播神经网络形成深度置信网络对字符图片数据进行识别,并且使用CUDA在图形处理器(GPU)上进行并行计算来完成识别过程。实验结果表明,使用该方法后,在不降低识别精度的情况下手写字符识别的速度大幅提升。
By combining Compute Unified Device Architecture (CUDA) and Deep Belief Network (DBN), this paper proposes a method to handle the recognition of massive handwritten characters. This method integrates Restricted Boltzmann Machine (RBM) and back-propagation neural network to forming a DBN to recognize the characters. Moreover, the concurrent computation of CUDA is made in GPU to achieve the recognition. The results show that the proposed method can significantly improve the speed of recognition on handwritten characters without reducing the accuracy of recognition.
参考文献:
正在载入数据...
