深度学习在笔迹鉴定中的应用研究  

Application of deep learning in handwritten identification

作　　者：沈荣 黄晨 SHEN Rong;HUANG Chen(School of Intelligent Manufacturing,Sichuan University of Arts and Science,Dazhou 635000,China;Dazhou Public Security Bureau,Dazhou 635000,China)

机构地区：[1]四川文理学院智能制造学院,四川达州635000 [2]四川省达州市公安局,四川达州635000

出　　处：《电子设计工程》2020年第21期159-163,共5页Electronic Design Engineering

基　　金：四川文理学院2017年度智能制造产业技术开发研究重点项目(2017zn22003z)。

摘　　要：传统笔迹鉴定起源于“经验性”累积,科学基础尚需证实与加强,在司法实践中经常出现争议。对此,引入基于深度学习的目标检测算法,提出一种应用深度学习技术实现对笔迹识别的设计方案。实现过程中,建立并优化了全连接神经网络模型,并将模型移植到ROS(Robot Operating System)机器人系统上,实现了对手写体数字的识别应用。以MNIST数据集作为模拟检材,对手写体数字的识别正确率可达98.32%。证明深度学习技术可以有效挖掘图像信息,是替代传统“经验性”文件鉴定的手段,对拓宽法医学、鉴定科学研究方法具有启发意义和应用参考价值。Traditional handwriting identification data from"empirical"cumulation,scientific basis still need verify and enhance,disputes often arise in judicial practice.Therefore,introducing the object detection algorithm based on deep learning and come up with a deep learning technique is used to design handwriting recognition.In the process of realization,optimizing and building a fully connected neural network model and migrating models to ROS(Robot Operating System).At the same time,realizing the application of handwritten numeral recognition.With MNIST data set as the simulation material,the recognition accuracy of handwritten digits can reach 98.32%.It is proved that deep learning technology can excavate image information,which is an alternative to the traditional method of"empirical"document identification.It has enlightening significance and applied reference value to broaden the research methods of forensic medicine and identification science.

关 键 词：笔迹鉴定 深度学习 神经网络 TensorFlow框架 手写体数字 

分 类 号：TN99[电子电信—信号与信息处理]