Python深度学习卷积神经网络参数数量（第五章第2节 P107）

首先贴一下代码清单5-5 将猫狗分类的小型卷积神经网络实例化 代码

# 卷积网络实现
from keras import models
from keras import layers

model = models.Sequential()
model.add(layers.Conv2D(32,(3,3),activation = 'relu',input_shape = (150,150,3)))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(64,(3,3),activation = 'relu'))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(128,(3,3),activation = 'relu'))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(128,(3,3),activation = 'relu'))
model.add(layers.MaxPooling2D((2,2)))
# 平坦层
model.add(layers.Flatten())
model.add(layers.Dense(512,activation = 'relu'))
model.add(layers.Dense(1,activation = 'sigmoid'))
model.summary()

输出网络结构及参数如下

Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_9 (Conv2D)            (None, 148, 148, 32)      896       
_________________________________________________________________
max_pooling2d_9 (MaxPooling2 (None, 74, 74, 32)        0         
_________________________________________________________________
conv2d_10 (Conv2D)           (None, 72, 72, 64)        18496     
_________________________________________________________________
max_pooling2d_10 (MaxPooling (None, 36, 36, 64)        0         
_________________________________________________________________
conv2d_11 (Conv2D)           (None, 34, 34, 128)       73856     
_________________________________________________________________
max_pooling2d_11 (MaxPooling (None, 17, 17, 128)       0         
_________________________________________________________________
conv2d_12 (Conv2D)           (None, 15, 15, 128)       147584    
_________________________________________________________________
max_pooling2d_12 (MaxPooling (None, 7, 7, 128)         0         
_________________________________________________________________
flatten_3 (Flatten)          (None, 6272)              0         
_________________________________________________________________
dense_5 (Dense)              (None, 512)               3211776   
_________________________________________________________________
dense_6 (Dense)              (None, 1)                 513       
=================================================================
Total params: 3,453,121
Trainable params: 3,453,121
Non-trainable params: 0

1. 对于卷积层而言，假设卷积核尺寸为S* S， 滤器数量为F(即输出通道数)， 存在偏置项的情况下，则固定为1，输入通道数为C，则参数数量计算公式= (S*S*C + 1) * F； 
2. 对于池化层、flatten、dropout操作而言，是不需要训练参数的，所以param为0；
3. 对于全连接层而言，假设输入的列向量大小为Fl(即平坦层)，输出的列向量大小为L，若开启bias，则param计算公式为=Fl *L + L

所以：

第一层卷积层参数：896 =([3*3]*3+1)*32，中括号外边的3为通道数为3，32为输出通道数；

第二层卷积层参数：18496=([3*3]*32+1)*64;

第三层卷积层参数：73856=([3*3]*64+1)*128;

第四层卷积层参数：147584=([3*3]*128+1)*128;

平坦层后的全连接层参数：3211776=6272*512+512