tensorflow进行MNIST手写数字识别-CNN

使用两个卷积层，两个全连接层。

准确率0.99

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

mnist = input_data.read_data_sets('MNIST_data', one_hot=True)

batch_size = 100 #每个批次大小
n_batch = mnist.train.num_examples // batch_size#批次数量

#初始化权值
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev = 0.1)
    return tf.Variable(initial)

#初始化偏置值
def biase_variable(shape):
    initial = tf.constant(0.1, shape = shape)
    return tf.Variable(initial)

#卷积层
def conv2d(x, W):
    #x 是tensor [batch, in_height, in_width, in_channels]
    #W是滤波器 [filter_height, filter_width, in_channels, out_channels]
    #stride步长 strides[0] = strides[3] = 1, strides[1]代表x方向的步长，strides[2]代表y方向的步长,[1, x, y, 1]
    #padding两种方式，分别是SAME和VALID，SAME会补0，VALID不会补0
    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]
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides = [1, 2, 2, 1], padding = 'SAME')

x = tf.placeholder(tf.float32, shape = [None, 784])
y = tf.placeholder(tf.float32, shape = [None, 10])

#改变x的格式转换为4D向量
x_image = tf.reshape(x, [-1, 28, 28, 1]) #[batch, in_height, in_width, in_channels] in_channels黑白图片为1， 彩色图片为3

#初始化第一个卷积层的权值和偏置值
W_conv1 = weight_variable([5, 5, 1, 32])#5*5的采样窗口， 32个卷积核从1个平面抽取特征
b_conv1 = biase_variable([32])#每一个卷积核一个偏置值

#把x_image和权值向量进行卷积，再加上偏置值，然后应用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  第一次池化后图片变为14*14

#初始化第二个卷积层的权值和偏置
W_conv2 = weight_variable([5, 5, 32, 64])#5*5的采样窗口， 64个卷积核从32个平面抽取特征
b_conv2 = biase_variable([64])#每一个卷积核一个偏置值

#把h_pool1和权值向量进行卷积，再加上偏置值，然后应用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    #第二次池化后图片变为7*7

#初始化第一个全连接层的权值
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 = biase_variable([1024])

#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])
#求第一个全连接层的输出
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)

#keep_prob表示神经元的输出概率
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 = biase_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))
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)#用优化器进行训练

#计算准确率
correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    writer = tf.summary.FileWriter('logs/',sess.graph)
    for epoch in range(30):#训练30个周期
        for batch in range(n_batch):
            batch_x, batch_y = mnist.train.next_batch(batch_size)
            sess.run(train_step, feed_dict={x:batch_x, y:batch_y, keep_prob:0.7})

        acc = sess.run(accuracy, feed_dict={x:mnist.test.images[0:1000], y:mnist.test.labels[0:1000], keep_prob:1.0})#使用测试集进行评测准确率
        print('Iter',epoch,'Accuracy:',acc)