神经网络（model.summary())模型层的转换与层参数详解

简单的卷积

from keras import layers 
from keras import models
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1))) 
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(64, (3, 3), activation='relu'))
model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))
model.summary()

对应的架构

Layer (type)                  Output                Shape Param #
=================================================================
conv2d_1 (Conv2D)              (None, 26, 26, 32)       320 
_________________________________________________________________
max_pooling2d_1 (MaxPooling2D) (None, 13, 13, 32)        0 
_________________________________________________________________
conv2d_2 (Conv2D)              (None, 11, 11, 64)      18496 
_________________________________________________________________
max_pooling2d_2 (MaxPooling2D) (None, 5, 5, 64)          0 
_________________________________________________________________
conv2d_3 (Conv2D)              (None, 3, 3, 64)        36928
_________________________________________________________________
flatten_1 (Flatten)            (None, 576)               0 
_________________________________________________________________
dense_1 (Dense)                (None, 64)              36928 
_________________________________________________________________
dense_2 (Dense)                (None, 10)               650 
=================================================================
Total params: 93,322
Trainable params: 93,322

卷积后的输出计算

conv2d_1 (Conv2D) 输入为(28,28,1)，卷积核尺寸为(3,3)，过滤器个数为32，步长默认(1, 1)，所以:

Output：      (输入尺寸-卷积核尺寸+2xpadding)/步长+1
		   	  (28-3+2x0)/1+1=26--->（26,26,32）
Shape Param： 卷积核尺寸*2 x 通道数 x 过滤器数+过滤器数
               W=3*2x1x32=288,
			   B=32,
			   Param=288+32=B+W=320

池化后的输出计算

conv2d_1 (Conv2D)为(26,26,32)，经过池化大小为(2,2)，步长默认: If None, it will default to pool_size.

输出尺寸： (输入尺寸-池化尺寸)/步长+1
		   (26-2)/2+1=13---->(13,13,32)

全连接层

1.卷积后的全连接成的参数计算
全连接的前层conv2d_3 (Conv2D) (None, 3, 3, 64)
全连接层的神经元数 64

Shape Param：前层的尺寸*2x前层的核数x全连接层的神经元数+全连接层的神经元数
		     3*2 x 64 x 64 + 64= 36928

2.隐藏层的下一层全连接成的参数计算
前层神经元数：64
当前神经元数：10

Shape Param：前层神经元数 x 当前神经元数 + 当前神经元数
					64 x 10 + 10=650

Total params

为所以参数的总和：320 + 18496 +36928 + 36928 +650 = 93322