深度卷积神经网络下选票系统智能化识别研究与实现  

Research and implementation of intelligent ballot recognition system based on deep convolution neural network

作　　者：陈一凡 彭程[1] 韩啸 刘霆 CHEN Yifan;PENG Cheng;HAN Xiao;LIU Ting(Chengdu Institute of Computer Application,Chinese Academy of Sciences,Chengdu Sichuan 610041,China;University of Chinese Academy Sciences,Beijing 100049,China)

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]中国科学院大学,北京100049

出　　处：《计算机应用》2019年第S02期85-90,共6页journal of Computer Applications

摘　　要：针对传统选票识别系统识别率低、效率不够高的问题,提出一种基于深度卷积神经网络的高准确率、低误识率的选票手写符自动识别系统。将图像处理方法与数据增强方法融合后引入卷积神经网络并对传统网络结构进行改进,构建出选票手写符号识别模型,形成一个高可靠性、高精度、高效率的选票识别系统。在各个会议、选举中采集大量数据集,通过不同模型的训练和测试结果表明,相较于传统图像处理方法、BP神经网络学习算法等现有方式,该方法具有更高的准确率,且能准确识别非期待图案的无意义图案,识别准确度达到99%以上。该手写符识别系统已应用于计票通项目,且可以推广应用到满足不同任务,多种型号的会议系统中。Aiming at the low recognition rate and low efficiency of traditional ballot identification system,this paper proposed an automatic handwritten character recognition system with high accuracy and low misrecognition rate based on deep convolutional neural network.The image processing method and data enhancement method are merged into the convolutional neural network.The traditional network structure was also improved to construct ballot handwritten character recognition model,forming a ballot recognition system with highly reliability,high-precision and high efficiency.A large number of data sets are collected in various conferences and elections.The training and test results of different models show that the method has higher accuracy than existing methods such as traditional image processing methods and BP neural network learning algorithms.The system can also accurately identify non-meaningful patterns and non-expected patterns,with recognition accuracy of over 99%.The handwritten character recognition system proposed in this paper has been applied to the ticket counting project,and can be applied to various conference systems and different tasks.

关 键 词：选票系统 手写字符 图像处理 数据增强 深度学习 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]