【原创深度学习与TensorFlow 动手实践系列 - 3】第三课：卷积神经网络 - 基础篇...

╰半橙微兮° 2022-01-13 06:43 171阅读 0赞

【原创 深度学习与TensorFlow 动手实践系列 - 3】第三课：卷积神经网络 - 基础篇

![4915-20170606182238793-1696892321.png][]

![4915-20170606182312122-465433004.png][]

**提纲：**

**1. 链式反向梯度传到**

**2. 卷积神经网络 - 卷积层**

**3. 卷积神经网络 - 功能层**

**4. 实例：卷积神经网络MNIST分类**

![4915-20170606182822981-255607361.png][]

**期待目标：**

**1. 清楚神经网络优化原理，掌握反向传播计算。**

**2. 掌握卷积神经网络卷积层的结构特点，关键参数，层间的连接方式。**

**3. 了解不同卷积神经网络功能层的作用，会进行简单的卷积神经网络结构设计。**

**4. 能够运行TensorFlow卷积神经网络 MNIST。 **

**![4915-20170606183327606-1922993717.png][]**

![4915-20170606183352247-90067783.png][]

![4915-20170606183539450-113793706.png][]

f(x, y, z) = (x + y) \* z

(3.00 + 1.00) \* -2.00 = -8.00

![4915-20170606183744793-1842559313.png][]

![4915-20170606183809106-244748674.png][]

![4915-20170606183835262-1875811102.png][]

![4915-20170606183905450-550965856.png][]

![4915-20170606185132340-44039097.png][]

![4915-20170606185159200-214890141.png][]

![4915-20170606185253497-293346188.png][]

![4915-20170606185313278-97956983.png][]

![4915-20170606185349934-1335766278.png][]

![4915-20170606185409168-568952517.png][]

![4915-20170606185432153-2047163675.png][]

![4915-20170606185459418-1935966169.png][]

![4915-20170606185527528-1518519516.png][]

![4915-20170606185555247-2009175717.png][]

![4915-20170606185620981-2136767729.png][]

![4915-20170606185651512-1109902764.png][]

![4915-20170606185729325-2060073329.png][]

![4915-20170606185802590-1051759907.png][]

1. 什么是卷积层

2. 有什么组成

3. 受什么影响，有何特点

4. 怎么工作

![4915-20170606190007481-1757921516.png][]

**卷积层**

**1. 卷积神经网络基本结构**

**2. 由多个卷积核组合形成**

**3. 每个卷积核同输入数据卷积运算，形成新的特征“图”**

**![4915-20170606190412293-822608918.png][]**

![4915-20170606190428434-1169930514.png][]

![4915-20170606190519997-1403096324.png][]

![4915-20170606190552106-1883636975.png][]

![4915-20170606190615965-436893673.png][]

![4915-20170606190646606-782514016.png][]

![4915-20170606190732153-1832465901.png][]

![4915-20170606190814231-201456980.png][]

![4915-20170606190843512-1956397616.png][]

![4915-20170606190905012-404644239.png][]

![4915-20170606191017418-214393682.png][]

![4915-20170606191047043-1047959703.png][]

![4915-20170606191113512-1718327301.png][]

![4915-20170606191130247-991100735.png][]

![4915-20170606191152434-1588379893.png][]

![4915-20170606191209653-484693156.png][]

![4915-20170606191223325-430024100.png][]

![4915-20170606191249387-728671231.png][]

![4915-20170606191307497-273774745.png][]

**非线性激励层（None-linear activation layer）：卷积是线性运算，增加非线性描述能力**

![4915-20170606191357247-1298666245.png][]

**降维：特征图稀疏，减少数据运算量，保持精度**

**池化层（Pooling Layer）- 降维：**

**1. 数据降维，方便计算，存储（max，ave）**

**2. 池化过程中，每张特征图单独降维**

**![4915-20170606191815856-1675941327.png][]**

** 归一化：特征的scale保持一致。**

![4915-20170606191921512-2080404901.png][]

![4915-20170606191935450-1636805526.png][]

![4915-20170606192001403-2039010732.png][]

![4915-20170606192019528-1571044405.png][]

![4915-20170606192032465-590123117.png][]

![4915-20170606192053637-78681794.png][]

![4915-20170606192110434-147136284.png][]

![4915-20170606192127372-1731409320.png][]

![4915-20170606192221075-1822314100.png][]

![4915-20170606192331028-1506091804.png][]

**卷积神经网络（CNN） - 基础 演示代码：**

**https://github.com/wiibrew/DeepLearningCourseCodes**

**https://github.com/wiibrew/DeepLearningCourseCodes/blob/master/course\_3\_tf\_mnist\_cnn.py**

**https://github.com/wiibrew/DeepLearningCourseCodes/blob/master/course\_3\_tf\_mnist\_cnn.ipynb**

**卷积神经网络运行情况：**

![4915-20170609105056387-1095881789.png][]

转载于:https://www.cnblogs.com/licheng/p/6952997.html

[4915-20170606182238793-1696892321.png]: /images/20220113/d394b4a748564bf4ac130a24b579e951.png
[4915-20170606182312122-465433004.png]: /images/20220113/b3db06af25f745ffa1bf087f122b7291.png
[4915-20170606182822981-255607361.png]: /images/20220113/363a5cd44ac94d21b0a53b398cd71596.png
[4915-20170606183327606-1922993717.png]: /images/20220113/e0645bc812324fb5bd8e71cb92a2067c.png
[4915-20170606183352247-90067783.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183352247-90067783.png
[4915-20170606183539450-113793706.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183539450-113793706.png
[4915-20170606183744793-1842559313.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183744793-1842559313.png
[4915-20170606183809106-244748674.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183809106-244748674.png
[4915-20170606183835262-1875811102.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183835262-1875811102.png
[4915-20170606183905450-550965856.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606183905450-550965856.png
[4915-20170606185132340-44039097.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185132340-44039097.png
[4915-20170606185159200-214890141.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185159200-214890141.png
[4915-20170606185253497-293346188.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185253497-293346188.png
[4915-20170606185313278-97956983.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185313278-97956983.png
[4915-20170606185349934-1335766278.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185349934-1335766278.png
[4915-20170606185409168-568952517.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185409168-568952517.png
[4915-20170606185432153-2047163675.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185432153-2047163675.png
[4915-20170606185459418-1935966169.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185459418-1935966169.png
[4915-20170606185527528-1518519516.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185527528-1518519516.png
[4915-20170606185555247-2009175717.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185555247-2009175717.png
[4915-20170606185620981-2136767729.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185620981-2136767729.png
[4915-20170606185651512-1109902764.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185651512-1109902764.png
[4915-20170606185729325-2060073329.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185729325-2060073329.png
[4915-20170606185802590-1051759907.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606185802590-1051759907.png
[4915-20170606190007481-1757921516.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190007481-1757921516.png
[4915-20170606190412293-822608918.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190412293-822608918.png
[4915-20170606190428434-1169930514.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190428434-1169930514.png
[4915-20170606190519997-1403096324.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190519997-1403096324.png
[4915-20170606190552106-1883636975.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190552106-1883636975.png
[4915-20170606190615965-436893673.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190615965-436893673.png
[4915-20170606190646606-782514016.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190646606-782514016.png
[4915-20170606190732153-1832465901.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190732153-1832465901.png
[4915-20170606190814231-201456980.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190814231-201456980.png
[4915-20170606190843512-1956397616.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190843512-1956397616.png
[4915-20170606190905012-404644239.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606190905012-404644239.png
[4915-20170606191017418-214393682.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191017418-214393682.png
[4915-20170606191047043-1047959703.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191047043-1047959703.png
[4915-20170606191113512-1718327301.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191113512-1718327301.png
[4915-20170606191130247-991100735.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191130247-991100735.png
[4915-20170606191152434-1588379893.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191152434-1588379893.png
[4915-20170606191209653-484693156.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191209653-484693156.png
[4915-20170606191223325-430024100.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191223325-430024100.png
[4915-20170606191249387-728671231.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191249387-728671231.png
[4915-20170606191307497-273774745.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191307497-273774745.png
[4915-20170606191357247-1298666245.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191357247-1298666245.png
[4915-20170606191815856-1675941327.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191815856-1675941327.png
[4915-20170606191921512-2080404901.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191921512-2080404901.png
[4915-20170606191935450-1636805526.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606191935450-1636805526.png
[4915-20170606192001403-2039010732.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192001403-2039010732.png
[4915-20170606192019528-1571044405.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192019528-1571044405.png
[4915-20170606192032465-590123117.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192032465-590123117.png
[4915-20170606192053637-78681794.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192053637-78681794.png
[4915-20170606192110434-147136284.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192110434-147136284.png
[4915-20170606192127372-1731409320.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192127372-1731409320.png
[4915-20170606192221075-1822314100.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192221075-1822314100.png
[4915-20170606192331028-1506091804.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170606192331028-1506091804.png
[4915-20170609105056387-1095881789.png]: https://images2015.cnblogs.com/blog/4915/201706/4915-20170609105056387-1095881789.png