摘要：这次我毕设的课题为:基于深度学习的图像语义分割算法研究。在这次的毕设中我主要用到了以下的工具:Virtual Box虚拟机caffe,segnet以及python几大主流程序。

在近几年中,深度学习在各个领域中的发展都非常迅速,计算机视觉也就是人脸识别等的发展也是与日俱增。深度学习的卷积神经网络已经成为大家公认的在计算机视觉识别分析领域中起到了至关重要的作用。与此同时，在我们的日常生活中，这一项技术也时不时能够出现在我们的身边，从人脸识别到声控再到车载显示仪还是对食品的监控以及分析，他都起到了至关重要的作用。图像语义分割是大家所公认的在计算机深度学习领域中作为一种基础算法最基本的也是最重要的一个部分。图像语义分割的质量直接决定了后续算法所使用的基本素材，因此它在整个过程中起到了至关重要的作用。因此如何能够将这个算法的错误率降到最低是我们最需要重视的问题。

在这次毕设的研究过程中,最大的难点是如何解决在编程过程中遇到的种种报错,以及如何排除程序报错的问题。在这次的毕设过程中,大多数报错都是因为程序中的配置文件路径错误或者是程序中对GPU以及CPU类型定义错误造成,还有少许则是PYTHON的相关数据库缺失,需要下载造成。

在师生的共同努力以及对文献资料的查阅后克服了种种困难,最后实现了对图片实行图像分割以及识别的功能。

关键词： 图像语义分割；卷积神经网络；CAFFE；SEGNET；PYTHON

image semantic segmentation based on deep learning

Abstract: This time, I set up a topic for the study of image semantic segmentation based on deep learning. In this set up, I mainly used the following tools: Virtual Box virtual machine caffe, segnet and python several mainstream programs.

In recent years, the development of deep learning is very fast, and the development of computer is fast.vision has also followed a wave of development. Deep convolutional neural networks have been proven to be very effective tools and methods in the field of computer vision. At the same time, in daily life, whether it is the detection of items, or video surveillance data analysis is inseparable from computer vision. The semantic segmentation of images is considered as the most basic algorithm in the field of computer vision. The semantic segmentation of images directly determines the classification or recognition of subsequent algorithms. Therefore, the implementation and application of an effective image semantic segmentation algorithm have very important practical significance.

In this research process, the biggest difficulty is how to solve all kinds of errors encountered in the programming process, and how to eliminate the problem of program error. In this process, most of the errors are caused by incorrect configuration file paths in the program or errors in the GPU and CPU type definitions in the program. There are also a few PYTHON related databases that are missing and need to be downloaded.

After the joint efforts of the teachers and students and the review of the documents, the difficulties were overcome. Finally, the function of image segmentation and recognition on the pictures was realized.

Keywords: Image semantic segmentation; Convolutional neural network; CAFFE; SEGNET; PYTHON

目录

摘要 i

Abstract i

目录 iii

1 绪论 1

1.1 什么是卷积神经网络 2

1.1.1 卷积神经网络的构成与实际使用 2

1.1.2 ImageNet图像分类 5

1.2 图像语义分割 6

1.2.1 图像分割技术常用方法 6

1.2.2 图像语义分割算法 7

1.2.3 前端 8

1.2.4 后端 9

1.2.5 小结 10

1.3 修改跨层融合结构 11

1.4 文本组织和结构 13

2 分析 14

2.1 环境安装 14

2.2 SEGNET介绍 15

2.3 实验结果 16