基于深度学习的图像识别与文字推荐系统的设计与实现

发布时间：2018-03-07 23:07

本文选题：深度学习　切入点：卷积神经网络　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：深度学习(DL,Deep Learning)是机器学习(ML,Machine Learning)的一个重要方法和研究方向,属于人工智能(AI,Artificial Intelligence)领域的重要分支。随着大数据技术的发展,深度学习迎来了又一个快速发展的时期,这也使得深度学习理论与算法研究焕发新的活力。卷积神经网络(CNN,Convolutional Neural Network)作为深度学习模型的代表,是模拟视觉系统层次化的工作模式,在人工神经网络的基础上构建具有层次化结构的人工网络模型。其局部感知、层次结构化等特点在处理图像识别问题上具有巨大优势,在现代模式识别领域获得了广泛应用。本文在整理与总结国内外深度学习的基本理论成果与在工程上的应用现状,并对卷积神经网络结构分析的基础上,结合Word2Vec与TensorFlow深度学习框架,开发了图像识别与文字推荐系统,以工程应用为背景对其理论成果进行研究。本文主要进行了以下几项工作:整理国内外深度学习的研究成果,并对深度学习的背景与应用进行总结;分析卷积神经网络与Word2Vec的结构与基本原理,并对理解网络模型所需的基本算法进行了介绍;设计本文的图像识别与文字推荐系统,并以经典CNN网络结构为基础设计基于本文推荐的卷积神经网络结构;进行数据集的准备、深度学习框架的搭建及本文模型训练工作,并实现本文图像识别与文字推荐系统;通过以上工作,本文从工程项目应用的角度验证了深度学习在图像识别与自然语言处理问题上的优势。
[Abstract]:Deep learning is an important method and research direction of machine learning, which belongs to the important branch of artificial intelligence. With the development of big data technology, deep learning has ushered in another period of rapid development. As a representative of depth learning model, convolution neural network (CNN) is a hierarchical working mode of simulating visual system. An artificial network model with hierarchical structure is constructed on the basis of artificial neural network. Its local perception and hierarchical structure have great advantages in image recognition. It has been widely used in the field of modern pattern recognition. In this paper, the basic theoretical achievements of deep learning at home and abroad and the present situation of application in engineering are summarized, and the network structure of convolutional neural network is analyzed. Based on the Word2Vec and TensorFlow deep learning framework, an image recognition and character recommendation system is developed, and its theoretical results are studied under the background of engineering application. The main work of this paper is as follows: sorting out the research results of deep learning at home and abroad. The background and application of deep learning are summarized, the structure and basic principle of convolutional neural network and Word2Vec are analyzed, and the basic algorithms for understanding the network model are introduced. Based on the classical CNN network structure, we design the convolutional neural network structure recommended in this paper, prepare the data set, build the deep learning framework and train the model in this paper, and realize the image recognition and text recommendation system in this paper. Through the above work, this paper verifies the advantages of deep learning in image recognition and natural language processing from the point of view of engineering project application.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP391.3

【参考文献】