LaTax卷积神经网络结构图（新）

已于 2022-10-31 10:57:19 修改
阅读量498 收藏
点赞数
文章标签： cnn python 人工智能
于 2022-09-29 12:03:32 首次发布
版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。
本文链接：https://blog.csdn.net/qq_45083082/article/details/127104492
版权
该博客展示了如何使用 LaTeX 和 TikZ 库创建一个卷积神经网络（CNN）的详细可视化图，包括多个卷积层、池化层、全连接层以及反向传播过程中的得分层和反卷积层。内容涵盖了从输入图像到最终分类的整个流程。
摘要由CSDN通过智能技术生成 
\documentclass[border=15pt, multi, tikz]{standalone}
\usepackage{import}
\subimport{../../layers/}{init}
\usetikzlibrary{positioning}
\usetikzlibrary{3d} %for including external image 

\def\ConvColor{rgb:yellow,5;red,2.5;white,5}
\def\ConvReluColor{rgb:yellow,5;red,5;white,5}
\def\PoolColor{rgb:red,1;black,0.3}
\def\DcnvColor{rgb:blue,5;green,2.5;white,5}
\def\SoftmaxColor{rgb:magenta,5;black,7}
\def\SumColor{rgb:blue,5;green,15}
\tikzstyle{copyconnection}=[ultra thick,every node/.style={sloped,allow upside down},draw={rgb:blue,4;red,1;green,1;black,3},opacity=0.7]
\newcommand{\copymidarrow}{\tikz \draw[-Stealth,line width =0.8mm,draw={rgb:blue,4;red,1;green,1;black,3}] (-0.3,0) -- ++(0.3,0);}
\begin{document}
\begin{tikzpicture}
\tikzstyle{connection}=[ultra thick,every node/.style={sloped,allow upside down},draw=\edgecolor,opacity=0.7]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Draw Layer Blocks
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\node[canvas is zy plane at x=0] (temp) at (-3,0,0) {\includegraphics[width=8cm,height=8cm]{cats.png}};
% conv1_1,conv1_2,%pool1
\pic[shift={(0,0,0)}] at (0,0,0) {RightBandedBox={name=cr1,caption=Conv,%
        ,fill=\ConvColor,bandfill=\ConvReluColor,%
        height=40,width={2},depth=40}};


\pic[shift={(4,0,0)}] at (cr1-east) {RightBandedBox={name=cr110,caption=Conv,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=40,width={2},depth=40}};

\pic[shift={(4,0,0)}] at (cr110-east) {RightBandedBox={name=cr11,caption=Conv,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=40,width={2},depth=40}};

\pic[shift={(4,0,0)}] at (cr11-east) {RightBandedBox={name=cr112,caption=Conv,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=40,width={2},depth=40}};

% conv3_1,conv3_2,pool3
\pic[shift={(3,0,0)}] at (cr112-east) {RightBandedBox={name=cr2,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={2},depth=6}};

\pic[shift={(0,0,0)}] at (cr2-east) {RightBandedBox={name=cr21,caption=Block1,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={2},depth=18}};

\pic[shift={(0,0,0)}] at (cr21-east) {RightBandedBox={name=cr22,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={8},depth=6}};

\pic[shift={(0,0,0)}] at (cr22-east) {Box={name=p2,caption=MaxPooling,%
        fill=\PoolColor,opacity=0.5,height=12,width=8,depth=12}};


\pic[shift={(3,0,0)}] at (p2-east) {RightBandedBox={name=cr3,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={2},depth=6}};

\pic[shift={(0,0,0)}] at (cr3-east) {RightBandedBox={name=cr31,caption=Block1,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={2},depth=18}};

\pic[shift={(0,0,0)}] at (cr31-east) {RightBandedBox={name=cr32,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={8},depth=6}};

\pic[shift={(0,0,0)}] at (cr32-east) {Box={name=p3,caption=MaxPooling,%
        fill=\PoolColor,opacity=0.5,height=12,width=8,depth=12}};


\pic[shift={(6,2.5,0)}] at (p3-east) {RightBandedBox={name=cr4,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=6,width={4},depth=6}};

\pic[shift={(0,0,0)}] at (cr4-east) {RightBandedBox={name=cr41,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=6,width={4},depth=18}};

\pic[shift={(0,0,0)}] at (cr41-east) {RightBandedBox={name=cr42,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=18,width={6},depth=6}};


\pic[shift={(6,-2.6,0)}] at (p3-east) {RightBandedBox={name=cr5,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={4},depth=6}};

\pic[shift={(0,0,0)}] at (cr5-east) {RightBandedBox={name=cr51,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={4},depth=18}};

\pic[shift={(0,0,0)}] at (cr51-east) {RightBandedBox={name=cr52,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={6},depth=18}};

\pic[shift={(7,5.8,0)}] at (p3-east) {RightBandedBox={name=cr6,%
      fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={6},depth=6}};

\pic[shift={(6,-7,0)}] at (p3-east) {RightBandedBox={name=cr7,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={6},depth=6}};
\pic[shift={(0,0,0)}] at (cr7-east) {Box={name=p4,caption=Block2,%
        fill=\PoolColor,opacity=0.5,height=12,width=6,depth=12}};


\pic[shift={(12,0,0)}] at (p3-east) {Ball={name=elt1,%
        fill=\SumColor,opacity=0.6,%
        radius=2.5,logo=$+$}};






\pic[shift={(10,0,0)}] at (cr41-east) {RightBandedBox={name=cr8,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=6,width={4},depth=6}};

\pic[shift={(0,0,0)}] at (cr8-east) {RightBandedBox={name=cr81,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=6,width={4},depth=18}};

\pic[shift={(0,0,0)}] at (cr81-east) {RightBandedBox={name=cr82,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
      height=18,width={6},depth=6}};


\pic[shift={(10,0,0)}] at (cr51-east) {RightBandedBox={name=cr9,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={4},depth=6}};

\pic[shift={(0,0,0)}] at (cr9-east) {RightBandedBox={name=cr91,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={4},depth=18}};

\pic[shift={(0,0,0)}] at (cr91-east) {RightBandedBox={name=cr92,%
       fill=\ConvColor,bandfill=\ConvReluColor,%
        height=18,width={6},depth=18}};

\pic[shift={(10.3,0,0)}] at (cr6-east) {RightBandedBox={name=cr10,%
      fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={6},depth=6}};

\pic[shift={(9,0,0)}] at (p4-east) {RightBandedBox={name=cr12,%
        fill=\ConvColor,bandfill=\ConvReluColor,%
        height=6,width={6},depth=6}};
\pic[shift={(0,0,0)}] at (cr12-east) {Box={name=p5,caption=Block2,%
        fill=\PoolColor,opacity=0.5,height=12,width=6,depth=12}};

\pic[shift={(11.1,0,0)}] at (elt1-east) {Ball={name=elt2,%
        fill=\SumColor,opacity=0.6,%
        radius=2.5,logo=$+$}};

\pic[shift={(4.3,0,0)}] at (elt2-east) {Box={name=p6,caption=AveragePooling,%
        fill=\PoolColor,opacity=0.5,height=46,width=8,depth=46}};
\pic[shift={(6.5,0,0)}] at (p6-east) {Box={name=p7,caption=GlobalAveragePooling,%
        fill=\PoolColor,opacity=0.5,height=2,width=2,depth=150}};

\pic[shift={(4.5,0,0)}] at (p7-east) {Box={name=sigmoid,caption=Sigmoid,%
       fill=\SoftmaxColor,%
        height=1,width=1,depth=4}}; 



\pic[shift={(0,2,0)}] at (cr110-north) {RightBandedBox={name=score8,%
        fill=\ConvColor,%
        height=0,width=0,depth=0}};
\pic[shift={(6.3,0,0)}] at (cr1-east) {Box={name=d1,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  


\pic[shift={(0,2,0)}] at (cr112-north) {RightBandedBox={name=score7,%
        fill=\ConvColor,%
        height=0,width=0,depth=0}}; 
\pic[shift={(6.3,0,0)}] at (cr11-east) {Box={name=d2,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};   



\pic[shift={(0,2.5,0)}] at (p2-north) {RightBandedBox={name=score6,%
        fill=\ConvColor,%
        height=0,width=0,depth=0}}; 
\pic[shift={(1,0,0)}] at (p2-east) {Box={name=d3,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  

\pic[shift={(0,2.5,0)}] at (p3-north) {RightBandedBox={name=score61,%
        fill=\ConvColor,%
        height=0,width=0,depth=0}};
\pic[shift={(1,0,0)}] at (p3-east) {Box={name=d31,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  

\pic[shift={(5,0,0)}] at (p3-east) {Box={name=d4,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  

\pic[shift={(3,0,0)}] at (elt1-east) {Box={name=d5,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  

\pic[shift={(2.4,0,0)}] at (d31-east) {Box={name=d6,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}};  
\pic[shift={(2.2,0,0)}] at (elt1-east) {Box={name=d7,%
        fill=\DcnvColor,%
        height=0,width=0,depth=0}}; 


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Draw connections
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\draw [connection]  (cr1-east)    -- node {\midarrow} (cr110-west);
\draw [connection]  (cr110-east)    -- node {\midarrow} (cr11-west);
\draw [connection]  (cr11-east)    -- node {\midarrow} (cr112-west);
\draw [connection]  (cr112-east)    -- node {\midarrow} (cr2-west);
\draw [connection]  (p2-east)    -- node {\midarrow} (cr3-west);
\draw [connection]  (p6-east)    -- node {\midarrow} (p7-west);
\draw [connection]  (p7-east)    -- node {\midarrow} (sigmoid-west);
\draw [connection]  (d31-east)    -- node {\midarrow} (d6-west);
\draw [connection]  (elt1-east)    -- node {\midarrow} (d7-west);
\draw [connection]  (elt2-east)    -- node {\midarrow} (p6-west);
\draw [connection]  (p3-east)    -- node {\midarrow} (d31-west);


\path (p3-east) -- (d4-west) coordinate[pos=0.68] (between3_6) ;
\draw [connection]  (between3_6)    -- node {\midarrow} (cr6-west-|between3_6) -- node {\midarrow} (cr6-west);
\draw [connection]  (cr6-east) -- node {\midarrow} (cr6-east -| elt1-north) -- node {\midarrow} (elt1-north);

\path (p3-east) -- (d4-west) coordinate[pos=0.68] (between3_4) ;
\draw [connection]  (between3_4)    -- node {\midarrow} (cr4-west-|between3_4) -- node {\midarrow} (cr4-west);
\draw [connection]  (cr42-east) -- node {\midarrow} (cr42-east -| elt1-north) -- node {\midarrow} (elt1-north);

\path (p3-east) -- (d4-west) coordinate[pos=0.68] (between3_5) ;
\draw [connection]  (between3_5)    -- node {\midarrow} (cr5-west-|between3_5) -- node {\midarrow} (cr5-west);
\draw [connection]  (cr52-east) -- node {\midarrow} (cr52-east -| elt1-north) -- node {\midarrow} (elt1-north);

\path (p3-east) -- (d4-west) coordinate[pos=0.68] (between3_7) ;
\draw [connection]  (between3_7)    -- node {\midarrow} (cr7-west-|between3_7) -- node {\midarrow} (cr7-west);
\draw [connection]  (p4-east) -- node {\midarrow} (p4-east -| elt1-north) -- node {\midarrow} (elt1-north);


\path (elt1-east) -- (d5-west) coordinate[pos=0.73] (between1_8) ;
\draw [connection]  (between1_8)    -- node {\midarrow} (cr8-west-|between1_8) -- node {\midarrow} (cr8-west);
\draw [connection]  (cr82-east) -- node {\midarrow} (cr82-east -| elt2-north) -- node {\midarrow} (elt2-north);

\path (elt1-east) -- (d5-west) coordinate[pos=0.73] (between1_9) ;
\draw [connection]  (between1_9)    -- node {\midarrow} (cr9-west-|between1_9) -- node {\midarrow} (cr9-west);
\draw [connection]  (cr92-east) -- node {\midarrow} (cr92-east -| elt2-north) -- node {\midarrow} (elt2-north);

\path (elt1-east) -- (d5-west) coordinate[pos=0.73] (between1_10) ;
\draw [connection]  (between1_10)    -- node {\midarrow} (cr10-west-|between1_10) -- node {\midarrow} (cr10-west);
\draw [connection]  (cr10-east) -- node {\midarrow} (cr10-east -| elt2-north) -- node {\midarrow} (elt2-north);

\path (elt1-east) -- (d5-west) coordinate[pos=0.73] (between1_12) ;
\draw [connection]  (between1_12)    -- node {\midarrow} (cr7-west-|between1_12) -- node {\midarrow} (cr12-west);
\draw [connection]  (p5-east) -- node {\midarrow} (p5-east -| elt2-north) -- node {\midarrow} (elt2-north);













\path (cr1-east) -- (cr110-west) coordinate[pos=0.45] (between1_110) ;
\draw [connection]  (between1_110) -- node {\midarrow} (score8-west-|between1_110) -- node {\midarrow} (score8-west);
\draw [connection]  (score8-east) -- node {\midarrow} (score8-east -| d1-north)-- node {\midarrow} (d1-north);


\path (cr11-east) -- (cr112-west) coordinate[pos=0.45] (between11_112) ;
\draw [connection]  (between11_112) -- node {\midarrow} (score8-west-|between11_112) -- node {\midarrow} (score7-west);
\draw [connection]  (score7-east) -- node {\midarrow} (score7-east -| d2-north)-- node {\midarrow} (d2-north);


\path (d2-east) -- (cr2-west) coordinate[pos=0.25] (between2_2) ;
\draw [connection]  (between2_2) -- node {\midarrow} (score6-west-|between2_2) -- node {\midarrow} (score6-west);
\draw [connection]  (score6-east) -- node {\midarrow} (score6-east -| d3-north)-- node {\midarrow} (d3-north);


\path (d3-east) -- (cr3-west) coordinate[pos=0.25] (between3_3) ;
\draw [connection]  (between3_3) -- node {\midarrow} (score61-west-|between3_3) -- node {\midarrow} (score61-west);
\draw [connection]  (score61-east) -- node {\midarrow} (score61-east -| d31-north)-- node {\midarrow} (d31-north);











%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%


    


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\end{tikzpicture}
\end{document}\grid

益益和仔仔
关注
基于图卷积神经网络的人工智能：一种新的图像识别技术
程序员光剑
06-27 4147
近年来，随着深度学习技术的快速发展，图像识别领域也取得了显著的进步。传统的图像识别方法主要依赖于特征提取和分类器模型，但这些方法在处理大规模图像时，效果越来越差。图卷积神经网络 (GCN) 的出现为图像识别带来了新的思路和方法。
颜色的RGB数字表示方法
wmnmtm的专栏
01-18 5908
RGB色光的混合是加色混合，参看《色彩混合》。关于十六进制，参看《二进制的基础知识》。 颜色的RGB值表示法 从物理光学试验中得出：红、绿、蓝三种色光是其他色光所混合不出来的。而这三种色光以不同比例的混合几乎可以得出自然界所有的颜色。 如红光与不同比例的绿光混合可以得出橙、黄、黄绿等色；红光与不同比例的蓝紫光混合可以得出品红、红紫、紫红蓝；紫光与不同比例的绿光混合可以得出绿蓝、青、青...
卷积神经网络的基本操作,卷积神经网络百度百科
shirley67269的博客
08-19 991
卷积神经网络（ConvolutionalNeuralNetwork,CNN）是一种前馈神经网络，它的人工神经元可以响应一部分覆盖范围内的周围单元，对于大型图像处理有出色表现。[1] 它包括卷积层(alternatingconvolutionallayer)和池层(poolinglayer)。卷积神经网络是近年发展起来，并引起广泛重视的一种高效识别方法。
CNN卷积神经网络（图解CNN）
WGS.
08-23 7万+
超详细的CNN卷积网络讲解，及经典网络LeNet复现
卷积神经网络（CNN）的整体框架及细节（详细简单）
热门推荐
小林学编程的博客
04-06 8万+
详细简单的介绍卷积神经网络（CNN)，让读者快速掌握！！！
典型卷积神经网络结构总结
没有最好，只有更好！
11-04 4873
典型卷积神经网络结构总结 LeNet：最早用于数字识别的CNN层关系如下： Data layer：1*28*28 Convolutional layer(5*5)：20*24*24 Pooling(2*2, stride 2)：20*12*12 Convolutional layer(5*5)：50*8*8 Pooling(2*2, stride 2)：50*4*4 Inner Product：50
CNN卷积神经网络图解概述
weixin_43731369的博客
04-01 5107
CNN卷积神经网络概述什么是神经网络神经网络的基本单位前馈神经网络卷积神经网络为什么要卷积神经网络什么是卷积一维卷积二维卷积卷积神经网络结构卷积层汇聚层全连接层 什么是神经网络 人工神经网络实质上是数学模型，这些模型主要是通过对人脑的神经元网络进行抽象，构建人工神经元，并按照一定拓扑结构来建立人工神经元之间的连接，来模拟生物神经网络. 神经网络的基本单位 神经元，是构成神经网络的基本单元，其主要是...
卷积神经网络训练流程图,卷积神经网络训练过程
ynca67269的博客
10-05 2075
border_mode可以是valid或者full，具体看这里说明：.conv2d#激活函数用tanh#你还可以在(Activation('tanh'))后加上dropout的技巧: (Dropout(0.5))(Convolution2D(4, 5, 5, border_mode='valid',input_shape=(1,28,28))) (Activation('tanh'))#第二个卷积层，8个卷积核，每个卷积核大小3*3。· 过拟合一个batch增加了模型的容量并验证我们可以达到的最低损失。
Python-PlotNeuralNet用Latex代码绘制精美神经网络结构图
08-11
"PlotNeuralNet"是一个独特的库，它允许用户使用LaTeX代码来创建高质量、详细的神经网络结构图。这个库由Haris Iqbal开发，其源代码可以在GitHub上找到，版本号为81757e5。以下是对这个库及其使用方法的详细阐述。 ...
卷积神经网络画图神器PlotNeuralNet（基于Latex）使用记录
weixin_41892813的博客
05-16 6237
PlotNeuralNet工具是一个基于Latex，同时可用Python交互的卷积神经网络画图工具 目录1、安装使用与样例2、基本layer, block or unit的Latex语句（1）层（2）层的组合（3）残差结构 1、安装使用与样例 Github下载地址 一个例子，具体的安装与使用介绍看Github： 2、基本layer, block or unit的Latex语句 Latex文本可在~\PlotNeuralNet\examples\fcn8s 下直接运行成功 （1）层 \documentcl
卷积神经网络结构原理详细介绍
12-20
详细介绍卷积神经网络的结构原理，并且根据一个详细的例子来说明CNN网络如何应用
神经网络结构图绘图软件,神经网络示意图怎么画
Supermen333的博客
08-12 728
BP算法的基本思想是：学习过程由信号正向传播与误差的反向回传两个部分组成；正向传播时，输入样本从输入层传入，经各隐层依次逐层处理，传向输出层，若输出层输出与期望不符，则将误差作为调整信号逐层反向回传，对神经元之间的连接权矩阵做出处理，使误差减小。经反复学习，最终使误差减小到可接受的范围。具体步骤如下：1、从训练集中取出某一样本，把信息输入网络中。2、通过各节点间的连接情况正向逐层处理后，得到神经网络的实际输出。3、计算网络实际输出与期望输出的误差。...
现代卷积神经网络经典架构图
weixin_45745314的博客
02-18 2307
https://zh-v2.d2l.ai/
深度学习之卷积神经网络
Small__明的博客
11-16 7978
深度学习之卷积神经网络 文章目录深度学习之卷积神经网络卷积神经网络模型结构图一、卷积层卷积核映射到卷积层的过程：局部连接和权值共享二、池化层三、Softmax层四、超参padding:补充边界，在边界补一圈0或1.Stride 步幅 卷积神经网络模型结构图 输入层（Input layer) 卷积层（convolution layer) 池化层（pooling layer) 输出层（全连接层+Softmax layer) 一、卷积层 1、提取特征。压缩提纯。 2、卷积层级之间的神经元是局部连接和权值共享
DL入门(1)：卷积神经网络(CNN)
KI的博客
01-16 7万+
CNN通常包括这几层：输入层（Input layer）、卷积层（Convolutional layer）、池化层（pooling layer）以及输出层（全连接层+softmax layer）。
卷积神经网络基本构造
Clement
04-04 4421
卷积计算 卷积运算实现垂直边缘检测 过滤器检测不同方向的边缘 Padding 输入图片维度信息是[6,6]，卷积核尺寸[3,3]，padding=1，步长stride=1，经过卷积运算之后输出图片维度信息也是[(6+2x1-3+1),(6+2x1-3+1)]，这样也就保持维度信息的一致性，不致于特征信息丢失。 计算公式：n = (n-f+2p)/s+1 ，n是指输入/输出维度信息，f是卷积...
【人工智能与机器学习】基于卷积神经网络CNN的猫狗识别
apple_52030329的博客
07-02 9005
卷积神经网络（Convolutional Neural Networks，简称CNN）是一种具有局部连接、权值共享等特点的深层前馈神经网络（Feedforward Neural Networks），是深度学习（deep learning）的代表算法之一，擅长处理图像特别是图像识别等相关机器学习问题，比如图像分类、目标检测、图像分割等各种视觉任务中都有显著的提升效果，是目前应用最广泛的模型之一。
一文精简介绍CNN神经网络
m0_67912929的博客
05-28 8828
随着卷积网络层数的增加，误差的逆传播过程中存在的梯 度消失和梯度爆炸问题同样也会导致模型的训练难以进行，甚至会出现随着网络深度的加深，模型在训练集上的训练误差会出现先降低再升高的现象。CNN的训练，也称神经网络的学习算法与经典BP网络是一样的，都属于随机梯度下降（SGD：Stochastic Gradient Descent），也称增量梯度下降，实验中用于优化可微分目标函数的迭代算法。在CSDN上2021-11-19发布的一篇深度学习的卷积神经网，内容整理的精简，移动，现在将其进行转载，并发布在公众号“
卷积神经网络的数学公式
最新发布
02-23
C#卷积神经网络（Convolutional Neural Network，CNN）是一种常用于图像识别计算机视觉任务的深度学习模型。它通过使用卷积层、池化层和全连接层等组件来提取图像特征并进行分类。 数学公式在描述CNN时起到了重要的...
评论
添加红包

请填写红包祝福语或标题

个

红包个数最小为10个

元

红包金额最低5元

当前余额3.43元 前往充值 >
需支付：10.00元
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付元
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明：

1.余额是钱包充值的虚拟货币，按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载，可以购买VIP、付费专栏及课程。

余额充值