04卷积神经网络
绪论
无处不在的卷积神经网络
分类、检索、检测、分割、人脸识别、图像生成、自动驾驶。。。
深度学习三部曲
- 搭建神经网络
- 找到一个合适的损失函数
- 找到一个合适的优化函数,更新参数
基本组成结构
卷积:对于两个实变函数的一种数学操作
二维卷积:图像处理中,图像以二维矩阵形式输入神经网络
池化:保留主要特征同时减少参数和计算量,防止过拟合,提高模型泛化能力
一般存在卷积层与卷积层之间,全连接层与全连接层之间
卷积神经网络典型结构
Alexnet
![04 - 卷积神经网络[01-08-50][20201025-231401660]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025231409247-609587590.jpg)
![04 - 卷积神经网络[01-15-07][20201025-23191818]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025231923202-289356793.jpg)
![04 - 卷积神经网络[01-17-22][20201025-23215540]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025232203781-1011085167.jpg)
![04 - 卷积神经网络[01-18-27][20201025-232341942]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025232346095-1440213462.jpg)
AlexNet分层解析
![]() |
![]() |
![]() |
|---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
ZFNet
网络结构与AlexNet相同
仅更改卷积层一感受野,卷积层3、4、5滤波器个数
VGG
一个更深的网络
GoogleNet
![]() |
![]() |
|---|---|
![]() |
![]() |
RetNet
![]() |
![]() |
|---|---|

![04 - 卷积神经网络[01-07-14][20201025-231246711]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025231254047-1629752627.jpg)
![04 - 卷积神经网络[01-21-30][20201025-234538598]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235118878-427291284.jpg)
![04 - 卷积神经网络[01-22-52][20201025-234700688]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235208776-51364883.jpg)
![04 - 卷积神经网络[01-22-55][20201025-23464963]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235158615-1540443718.jpg)
![04 - 卷积神经网络[01-23-13][20201025-234725548]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235226280-890487927.jpg)
![04 - 卷积神经网络[01-23-16][20201025-234731997]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235246580-1652373287.jpg)
![04 - 卷积神经网络[01-23-22][20201025-234741170]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235254942-1625203343.jpg)
![04 - 卷积神经网络[01-23-47][20201025-234823113]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235320123-2042979837.jpg)
![04 - 卷积神经网络[01-23-57][20201025-234831768]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235323606-704619489.jpg)
![04 - 卷积神经网络[01-24-53][20201025-235339321]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201025235347871-1360016346.jpg)
![04 - 卷积神经网络[01-32-09][20201026-003306792]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026003518939-940320202.jpg)
![04 - 卷积神经网络[01-34-04][20201026-003318705]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026003543968-980908351.jpg)
![04 - 卷积神经网络[01-35-00][20201026-003418760]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026003550124-1108322600.jpg)
![04 - 卷积神经网络[01-35-06][20201026-003423915]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026003554833-1791499473.jpg)
![04 - 卷积神经网络[01-45-15][20201026-004235229]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026004343328-164907505.jpg)
![04 - 卷积神经网络[01-46-14][20201026-004312444]](https://img2020.cnblogs.com/blog/1859480/202010/1859480-20201026004359098-1097671926.jpg)
浙公网安备 33010602011771号