本文選題：深度學習　切入點：卷積神經網絡　出處：《北京交通大學》2017年碩士論文　論文類型：學位論文

【摘要】：深度學習(DL,Deep Learning)是機器學習(ML,Machine Learning)的一個重要方法和研究方向,屬于人工智能(AI,Artificial Intelligence)領域的重要分支。隨著大數(shù)據技術的發(fā)展,深度學習迎來了又一個快速發(fā)展的時期,這也使得深度學習理論與算法研究煥發(fā)新的活力。卷積神經網絡(CNN,Convolutional Neural Network)作為深度學習模型的代表,是模擬視覺系統(tǒng)層次化的工作模式,在人工神經網絡的基礎上構建具有層次化結構的人工網絡模型。其局部感知、層次結構化等特點在處理圖像識別問題上具有巨大優(yōu)勢,在現(xiàn)代模式識別領域獲得了廣泛應用。本文在整理與總結國內外深度學習的基本理論成果與在工程上的應用現(xiàn)狀,并對卷積神經網絡結構分析的基礎上,結合Word2Vec與TensorFlow深度學習框架,開發(fā)了圖像識別與文字推薦系統(tǒng),以工程應用為背景對其理論成果進行研究。本文主要進行了以下幾項工作:整理國內外深度學習的研究成果,并對深度學習的背景與應用進行總結;分析卷積神經網絡與Word2Vec的結構與基本原理,并對理解網絡模型所需的基本算法進行了介紹;設計本文的圖像識別與文字推薦系統(tǒng),并以經典CNN網絡結構為基礎設計基于本文推薦的卷積神經網絡結構;進行數(shù)據集的準備、深度學習框架的搭建及本文模型訓練工作,并實現(xiàn)本文圖像識別與文字推薦系統(tǒng);通過以上工作,本文從工程項目應用的角度驗證了深度學習在圖像識別與自然語言處理問題上的優(yōu)勢。
[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.
【學位授予單位】：北京交通大學
【學位級別】：碩士
【學位授予年份】：2017
【分類號】：TP391.41;TP391.3

【參考文獻】

相關期刊論文 前2條

1 楊志義;朱婭婷;蒲勇;;基于統(tǒng)一計算設備架構技術的并行圖像處理研究[J];計算機測量與控制;2009年04期

2 沈艷軍,汪秉文,胡曉婭;多層前向神經網絡帶正則化因子的算法[J];系統(tǒng)工程與電子技術;2004年09期

，

本文編號：1581352