摘要：随着当今科技的不断发展，我们的社会正在迈向更加信息化、科技化的时代。如今在这个信息爆炸的年代，如何利用好我们身边唾手可得的信息，将决定我们社会能否更加快速，高效地发展。信息分为很多种，例如声音信息，文字信息及实物信息等等，而这些信息中最为常见的应该就是图像信息了，因此我们想利用好这些最常见的图像信息就要研究发展目标图像的识别算法及系统。现在无论是发展得如火如荼的无人驾驶技术亦或是AR技术都十分依赖图像识别技术。

现在的图像目标识别系统主要由图像中的目标分割，目标提取，目标分类三部分组成。因此本论文将以这三个部分为基础，再结合上卷积神经网络来进行图像目标识别的研究,最后使用Fast R-CNN等技术完成一个简单的目标识别系统。

关键词： 深度学习；人工智能；图像分割；目标识别；卷积神经网络

Target Recognition Algorithm Based on Deep Learning

Abstract: With the development of science and technology, our society is moving toward  to an era of more information and technology. In this era of information explosion, how to make good use of information we can get everywhere will determine whether our society can develop more quickly and efficiently. There are many types of information, such as sound information, text information and physical information. But the most common of these information should be image information. Therefore, if we want to use these most common image information, we must study the development of target image recognition systems. Now whether the fully automatic driverless car or AR technology is dependent on image recognition technology.

The current image target recognition system is consisted of three parts: target segmentation in the image, target extraction, and target classification. Therefore, this dissertation will use these three parts as the foundation and then combine the upper convolutional neural network to study the image target recognition. Finally I will finish a simple target recognition system using the Fast R-CNN system. 

Keywords: deep learning; artificial intelligence; image segmentation; target recognition; Convolution neural network

目录

摘要 i

Abstract i

目录 ii

1 绪论 1

1.1 研究背景 1

1.2 国内外研究现状 1

1.3 主要设计内容及具体工作 2

1.4 本文结构 2

2 相关研究分析 3

2.1 图像分割算法研究 3

2.1.1 选择性搜索算法 3

2.1.2 边缘盒检测算法 7

2.2 深度学习技术 9

2.3 深度卷积神经网络 10

3 目标检测算法综述 11

3.1