CNN 训练千年战争最低cost表很小,但是准确率很低是为什么

匿名用户不能发表回复!|您所在位置: &
&nbsp&&nbsp&nbsp&&nbsp
基于深度学习的图像超分辨率重建研究.doc 64页
本文档一共被下载:
次 ,您可全文免费在线阅读后下载本文档。
下载提示
1.本站不保证该用户上传的文档完整性,不预览、不比对内容而直接下载产生的反悔问题本站不予受理。
2.该文档所得收入(下载+内容+预览三)归上传者、原创者。
3.登录后可充值,立即自动返金币,充值渠道很便利
基于深度学习的图像超分辨率重建研究
你可能关注的文档:
毕业设计(论 文)
班级5133117
学生姓名 指导教师 日 基于深度学习的图像超分辨率重建研究
本文首先简要介绍人工神经网络的发展历程,然后介绍深度学习在计算机视觉方面的应用。然后介绍神经网络的一些理论知识,最后介绍深度学习中的卷积神经网络(CNN, Convolutional Neural Network)。
本文研究如何利用卷积神经网络实现超分辨率重建。卷积神经网络分为三层结构,第一层的作用是特征块的提取和表示,第二层的作用是非线性映射,第三层的作用是重建出高分辨率图像。本文首先将一个图像降采样再双三次插值作为低分辨率图像,作为卷积神经网络的输入,而高分辨率图像作为卷积神经网络的输出,利用卷积神经网络建立低分辨率,高分辨率之间的映射。最后针对该模型进行改进,再加入一层作为特征提取。最后利用深度学习框架TensorFlow实现上述模型。最后研究快速超分辨率重建模型,并针对模型层数和过滤器大小进行改进,与先前实验做比对。
关键字:超分辨率重建,卷积神经网络,深度学习,TensorFlow Image Super-Resolution Using Deep learning
Author: Chu Wen-yu
Tutor: Zhang Kun
Artificial Neural Network because of its strong ability to learn, get rapid development of artificial intelligence, let the Artificial Neural Network become the research upsurge again. Deep learning has been widely used in computer vision, speech processing, natural language processing and so on. The super-resolution(SR) technique is designed to refactor a low-resolution image through a series of algorithms to reconstruct the corresponding high-resolution image. Currently, the method of frequency domain, Non-uniform image interpolation, Projection onto convex set(POCS), Maximum a posterior(MPA) and sparse matrix method are the more mature methods. This paper mainly researches the realization of super-resolution(SR) reconstruction using deep learning.
In this thesis, first is a brief introduction of the development of artificial neural network, then introduces the application of deep learning in computer vision. With that introduces some theoretical knowledge of neural network, and finally introduces the convolution neural network(CNN) in deep learning.
This article mainly researches how to use the convolution neural network(CNN) to get the super-resolution reconstruction. The convolution neural network contains three structures, the effect of the first layer is Patch extraction and representation, the second is the function of Non-linear mapping, the role of the third l
正在加载中,请稍后...6被浏览606分享邀请回答06 条评论分享收藏感谢收起0添加评论分享收藏感谢收起写回答7被浏览502分享邀请回答0添加评论分享收藏感谢收起

我要回帖

更多关于 svm训练的准确率 的文章

 

随机推荐