利用卷积神经网络实现数字手写体图像识别

利用卷积神经网络实现数字手写体图像识别(论文14000字,外文翻译)
摘要：数字手写图像识别是近年来研究的热点，具有广泛的应用前景,同时也是一个非常具有挑战性的课题。模式识别随着计算机的出现和人工智能的发展,迅速发展成一门学科。它所研究的理论和方法在很多学科和领域中得到广泛的重视,推动了人工智能系统的发展,扩大了计算机应用的可能性。卷积神经网络是近年来深度学习理论同人工神经网络结合的模式识别方法，目前已经成为了图像识别领域中的研究热点之一。
本文从卷积神经网络的基本概念和算法入手，深入研究卷积神经网络理论,开展数字手写体图像识别，并利用caffe软件包实现MNIST数据库上的数字手写图像识别，编写相应的界面读入数字图片进行处理。论文的主要工作如下：
（1）研究和总结国内外人工神经网络和卷积神经网络的研究现状，并介绍卷积神经网络的基本概念和原理，分析目前卷积神经网络的一些优缺点。
（2）传统的卷积神经网络是应用于手写数字识别研究的，本文则采用样本学习方法来构造有效的训练集，在windows系统下安装caffe软件包，并通过caffe软件包学习训练MNIST数据库中的数字手写体图像，并通过Matlab编写界面和程序，能够读入数字图片并识别数字手写体图像。
关键词：卷积神经网络；人工神经网络；图像识别；手写数字识别；模式识别；

Realization of Digital Handwritten Image Recognition by Convolution Neural Network
Abstract:Digital handwriting image recognition is a hot topic in recent years, with a wide range of application prospects, but also a very challenging subject. With the emergence of computer and the development of artificial intelligence, pattern recognition quickly developed into a discipline. The theory and method of its research have been paid more and more attention in many disciplines and fields, which promoted the development of artificial intelligence system and expanded the possibility of computer application. Convolution neural network is a pattern recognition method combining deep learning theory with artificial neural network in recent years. It has become one of the research hotspots in image recognition.
Based on the basic concepts and algorithms of convolution neural network, this paper studies the convolution neural network theory and develops digital handwritten image recognition. Using the caffe software package to achieve MNIST database digital handwriting image recognition, write the appropriate interface to read the digital image for processing. The main work of the paper is as follows:
(1) This paper studies and summarizes the research status of artificial neural network and convolution neural network both at home and abroad, and introduces the basic concepts and principles of convolution neural network. It also analyzes some advantages and disadvantages of convolution neural network.
(2) The traditional convolution neural network is applied to the study of handwritten numeral recognition. In this paper, the active sample learning method is used to construct the effective training set, and the caffe software package is installed under the windows system. And learn through the caffe package to train digital handwritten images in the MNIST database. And through Matlab programming interface and procedures, to read digital images and identify digital handwritten images.
Keywords: Convolution neural network; Artificial neural network; Image recognition;  Handwritten digital recognition;  Pattern recognition;
 

目  录
第一章 绪论    1
1.1 卷积神经网络研究现状    1
1.2 图像识别    1
1.2.1 模式识别与图像识别    1
1.2.2 图像识别的应用    2
1.3 手写数字识别    3
1.3.1 手写数字识别的技术现状    3
1.3.2 手写数字识别的一般方法    3
1.4 手写数字系统的概述    4
1.5 本文内容安排    5
第二章 卷积神经网络    5
2.1 神经网络    5
2.1.1 神经网络的概述    5
2.1.2 神经网络的模型结构    6
2.1.3 神经网络的学习方法    7
2.2 卷积神经网络    8
2.2.1 卷积神经网络的概述    8
2.2.2 卷积神经网络的模型    8
2.2.3 卷积神经网络的架构    9
第三章 MNIST数据集的训练    9
3.1 caffe搭建环境    9
3.2 MNIST数据集的训练    12
3.2.1  LeNet:MNIST分类模型的介绍    12
3.2.2 MNIST数据集的训练与测试    13
第四章 手写数字体图像的识别    14
4.1 数字手写体图像的输入    14
4.2 图像预处理    15
4.2.1 二值化    15
4.2.2归一化    16
4.3 特征提取    16
4.3 系统流程图以及大致工作    17
4.4实验结果    18
第五章 总结与展望    20
5.1 全文总结    20
5.2 未来展望    20
参考文献    22
致谢    24