1.传统神经网络问题
2.卷积操作
不同卷积核的表现
两种卷积操作
3.池化
两种池化操作
4.CNN结构
5.CNN的MNIST手写数字识别
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
#载入数据
mnist=input_data.read_data_sets("MNIST_data",one_hot=True)
#每个批次的大小
batch_size=100 #一次训练100张图
#计算共有多少批次
n_batch=mnist.train.num_examples//batch_size
#初始化权值
def weight_variable(shape):
initial=tf.truncated_normal(shape,stddev=0.1)#生成一个截断的正态分布
return tf.Variable(initial)
#初始化偏置值
def bias_variable(shape):
initial=tf.constant(0.1,shape=shape)
return tf.Variable(initial)
#卷积层
def conv2d(x,W):
#x:输入形状 [batch,in_height, in_width, in_channels]
#W filter:输入形状 [filter_height,filter_weight,in_channels,out_channels]
#strides[0]=strides[3]=1,strides[1]=x方向步长,strides[2]=y方向步长
#padding:"SAME"或"VALID"
return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding="SAME")
#池化层
def max_pool_2x2(x):
#ksize=[1,x,y,1] 核大小
return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
#定义两个placeholder
x=tf.placeholder(tf.float32,[None,784])
y=tf.placeholder(tf.float32,[None,10])
#改变x格式转为 x:输入形状 [batch,in_height, in_width, in_channels] 形式
x_image=tf.reshape(x,[-1,28,28,1])
#初始化第一个卷积层权值和偏置
W_conv1=weight_variable([5,5,1,32]) #5*5采样窗口,32个卷积核从1个平面抽取特征
b_conv1=bias_variable([32]) #每个卷积核一个偏置
#把x_image和权值向量进行卷积,再加上偏置,用relu激活函数
h_conv1=tf.nn.relu(conv2d(x_image,W_conv1)+b_conv1)
h_pool1=max_pool_2x2(h_conv1)#进行max_pooling
#初始化第二个卷积层权值和偏置
W_conv2=weight_variable([5,5,32,64]) #5*5采样窗口,64个卷积核从32个平面抽取特征
b_conv2=bias_variable([64]) #每个卷积核一个偏置
#把h_pool1和权值向量进行卷积,再加上偏置,用relu激活函数
h_conv2=tf.nn.relu(conv2d(h_pool1,W_conv2)+b_conv2)
h_pool2=max_pool_2x2(h_conv2)#进行max_pooling
#28*28的图片第一次卷积后为28*28 'SAME'补零, 第一次池化后为14*14,32个特征平面
#第二次卷积14*14的图片卷积后为14*14,第二次池化后为7*7 ,64个特征平面
#把池化层2的输出扁平化为1维
h_pool2_flat=tf.reshape(h_pool2,[-1,7*7*64])
#初始化第一个全连接层
W_fc1=weight_variable([7*7*64,1024])#上一层有7*7*64个神经元,全连接层有1024个神经元
b_fc1=bias_variable([1024]) #1024个
#求第一个全连接层输出
h_fc1=tf.nn.relu(tf.matmul(h_pool2_flat,W_fc1)+b_fc1)
#用keep_prob表示神经元输出概率
keep_prob=tf.placeholder(tf.float32)
h_fc1_drop=tf.nn.dropout(h_fc1,keep_prob)
#初始化第二个全连接层
W_fc2=weight_variable([1024,10])
b_fc2=bias_variable([10])
#计算输出
prediction=tf.nn.softmax(tf.matmul(h_fc1_drop,W_fc2)+b_fc2)
#交叉熵代价函数
cross_entropy=tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y,logits=prediction))
#使用Adam优化
train_step=tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
#结果存放在布尔列表中
correct_prediction=tf.equal(tf.argmax(y,1),tf.argmax(prediction,1))
#求准确率
accuracy=tf.reduce_mean(tf.cast(correct_prediction,tf.float32))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
for epoch in range(21):
for batch in range(n_batch):
batch_xs,batch_ys=mnist.train.next_batch(batch_size) #取下一批次
sess.run(train_step,feed_dict={x:batch_xs,y:batch_ys,keep_prob:0.7})
acc=sess.run(accuracy,feed_dict={x:mnist.test.images,y:mnist.test.labels,keep_prob:1.0})
print("Iter "+str(epoch)+" ,Testing Accuracy: "+str(acc))
输出结果(部分)
Iter 0 ,Testing Accuracy: 0.8724
Iter 1 ,Testing Accuracy: 0.9727
Iter 2 ,Testing Accuracy: 0.978
Iter 3 ,Testing Accuracy: 0.9823
Iter 4 ,Testing Accuracy: 0.9829
Iter 5 ,Testing Accuracy: 0.9839
Iter 6 ,Testing Accuracy: 0.9862
Iter 7 ,Testing Accuracy: 0.9877
Iter 8 ,Testing Accuracy: 0.9877
Iter 9 ,Testing Accuracy: 0.9882
Iter 10 ,Testing Accuracy: 0.9882
Iter 11 ,Testing Accuracy: 0.9894
Iter 12 ,Testing Accuracy: 0.9899
Iter 13 ,Testing Accuracy: 0.99
Iter 14 ,Testing Accuracy: 0.9895
Iter 15 ,Testing Accuracy: 0.9908
Iter 16 ,Testing Accuracy: 0.9901
Iter 17 ,Testing Accuracy: 0.9894
Iter 18 ,Testing Accuracy: 0.991