tensorflow使用批量归一化对CNN网络手写数字识别优化

批量归一化：

        由于在深层网络中，不同层的分布都不一样，会导致训练时出现饱和的问题。而批量归一化就是为了缓解这个问题提出的。而且在实际应用中，批量归一化的收敛非常快，并且具有很强的泛化能力，某种情况下完全可以替代正则化和弃权。

下面介绍如何在手写数字识别中使用批量归一化函数

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
from tensorflow.contrib.layers.python.layers import batch_norm

x_input=tf.placeholder(dtype=tf.float32,shape=[None,784])
y_input=tf.placeholder(dtype=tf.float32,shape=[None,10])
train=tf.placeholder(dtype=tf.bool)
x_image=tf.reshape(x_input,[-1,28,28,1]) #变成一张28*28像素的一个通道的图片

conv1=tf.Variable(tf.truncated_normal([3,3,1,64],mean=0,stddev=0.1)) #定义卷积核
convb1=tf.Variable(tf.zeros([64])+0.1)  #定义偏执
#卷积后的进行批量归一化
#################################
batch_norm_layer1=batch_norm(tf.add(tf.nn.conv2d(x_image,conv1,padding="SAME"),convb1),
                   decay=0.9,updates_collections=None,is_training=train)   #train=1表示训练，
#################################
#激活函数输出
convlayer1=tf.nn.relu(batch_norm_layer1)
poollayer1=tf.nn.max_pool(convlayer1,ksize=[1,2,2,1],strides=[1,2,2,1],padding="SAME") #池化

conv2=tf.Variable(tf.truncated_normal([3,3,64,32],mean=0,stddev=0.1))
convb2=tf.Variable(tf.zeros([32])+0.1)
######################################
batch_norm_layer2=batch_norm(tf.add(tf.nn.conv2d(poollayer1,conv2,padding="SAME"),convb2),
                             decay=0.9,updates_collections=None,is_training=train)
######################################
convlayer2=tf.nn.relu(batch_norm_layer2)
poollayer2=tf.nn.max_pool(convlayer2,ksize=[1,2,2,1],strides=[1,2,2,1],padding="SAME")

#print(poollayer2)  #得到平铺后的个数shape=(?, 7, 7, 32)
x_nn=tf.reshape(poollayer2,[-1,7*7*32]) #转化为可用于全连接神经网络的输入

w1=tf.Variable(tf.truncated_normal([7*7*32,512],mean=0,stddev=0.1))
b1=tf.Variable(tf.zeros([512])+0.1)
layer1=tf.nn.tanh(tf.add(tf.matmul(x_nn,w1),b1))   #分类

w2=tf.Variable(tf.truncated_normal([512,256],mean=0,stddev=0.1))
b2=tf.Variable(tf.zeros([256])+0.1)
layer2=tf.nn.tanh(tf.add(tf.matmul(layer1,w2),b2))

w3=tf.Variable(tf.truncated_normal([256,10],mean=0,stddev=0.1))
b3=tf.Variable(tf.zeros([10])+0.1)
layer3=tf.add(tf.matmul(layer2,w3),b3)

logit=layer3
loss=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(labels=y_input,logits=logit))
train_step=tf.train.GradientDescentOptimizer(0.1).minimize(loss)
precise=tf.reduce_mean(tf.cast(tf.equal(tf.argmax(logit,1),tf.argmax(y_input,1)),dtype=tf.float32))

mnist=input_data.read_data_sets("MNIST_data",one_hot=True)
batch_size=1000
n_batch_size=mnist.train.num_examples//batch_size

with tf.Session() as sess:
    init=tf.global_variables_initializer()
    sess.run(init)
    for i in range(10):
        for j in range(n_batch_size):
            x_train_input,y_train_input=mnist.train.next_batch(batch_size)
            sess.run(train_step,feed_dict={x_input:x_train_input,y_input:y_train_input,train:True})
            precise_train=sess.run(precise,{x_input:x_train_input,y_input:y_train_input,train:True})
            loss_p=sess.run(loss,{x_input:x_train_input,y_input:y_train_input,train:True})
            print(precise_train)
            print(loss_p)