CNN最大的特点在于卷积的权值共享结构，可以大幅减少神经网络的参数量，防止过拟合的同时又降低了神经网络模型的复杂度。在CNN中，第一个卷积层会直接接受图像像素级的输入，每一个卷积操作只处理一小块图像，进行卷积变化后再传到后面的网络，每一层卷积都会提取数据中最有效的特征。这种方法可以提取到图像中最基础的特征，比如不同方向的边或者拐角，而后再进行组合和抽象形成更高阶的特征。
总结一下，CNN的要点是局部连接（local Connection）、权值共享（Weight Sharing）和池化层（Pooling）中的降采样（Down-Sampling）。
from tensorflow.examples.tutorials.mnist import input_dataimport tensorflow as tf# 载入MNIST数据集，并创建默认的Interactive Session。mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)sess = tf.InteractiveSession()# 创建权重和偏置，以便重复使用。我们需要给权重制造一些随机的噪声来打破完全对称，比如截断的正态分布噪声，标准差设为0.1def weight_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial)def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial)# 创建卷积层、池化层，以便重复使用def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')def max_pool(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')# 定义输入的placeholderx = tf.placeholder(tf.float32, [None, 784])y_ = tf.placeholder(tf.float32, [None, 10])x_image = tf.reshape(x, [-1, 28, 28, 1])# 定义第一个卷积层W_conv1 = weight_variable([5, 5, 1, 32])b_conv1 = bias_variable([32])h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)h_pool1 = max_pool(h_conv1)# 定义第二个卷积层W_conv2 = weight_variable([5, 5, 32, 64])b_conv2 = bias_variable([64])h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)h_pool2 = max_pool(h_conv2)# 定义全连接层。由于第二个卷积层输出的tensor是7*7*64，我们使用tf.reshape函数对其进行变形W_fc1 = weight_variable([7*7*64, 1024])b_fc1 = bias_variable([1024])h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)# 为了减轻过拟合，下面使用一个Dropout层。通过一个placeholder传入keep_prob比率来控制的。在训练时，我们随机丢弃一部分节点# 的数据来减轻过拟合，预测时则保留全部数据来追求最好的预测性能。keep_prob = tf.placeholder(dtype=tf.float32)h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)# 最后我们将Dropout层的输出连接一个Softmax层，得到最后的概率输出W_fc2 = weight_variable([1024, 10])b_fc2 = bias_variable([10])y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)# 定义损失函数为cross entropy和优化器cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y_conv), reduction_indices=[1]))train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)# 定义评测准确率的操作correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))# 下面开始训练tf.global_variables_initializer().run()for i in range(20000): batch = mnist.train.next_batch(50) if i % 100 == 0: train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0}) print("Step %d, training accuracy %g" % (i, train_accuracy)) train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})print("test accuracy %g" % accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0}))# 载入MNIST数据集，并创建默认的Interactive Session。mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)sess = tf.InteractiveSession()# 创建权重和偏置，以便重复使用。我们需要给权重制造一些随机的噪声来打破完全对称，比如截断的正态分布噪声，标准差设为0.1def weight_variable(shape): initial = tf.truncated_normal(shape, stddev=0.1) return tf.Variable(initial)def bias_variable(shape): initial = tf.constant(0.1, shape=shape) return tf.Variable(initial)# 创建卷积层、池化层，以便重复使用def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')def max_pool(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')# 定义输入的placeholderx = tf.placeholder(tf.float32, [None, 784])y_ = tf.placeholder(tf.float32, [None, 10])x_image = tf.reshape(x, [-1, 28, 28, 1])# 定义第一个卷积层W_conv1 = weight_variable([5, 5, 1, 32])b_conv1 = bias_variable([32])h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)h_pool1 = max_pool(h_conv1)# 定义第二个卷积层W_conv2 = weight_variable([5, 5, 32, 64])b_conv2 = bias_variable([64])h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)h_pool2 = max_pool(h_conv2)# 定义全连接层。由于第二个卷积层输出的tensor是7*7*64，我们使用tf.reshape函数对其进行变形W_fc1 = weight_variable([7*7*64, 1024])b_fc1 = bias_variable([1024])h_pool2_flat = tf.reshape(h_pool2, [-1, 7*7*64])h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)# 为了减轻过拟合，下面使用一个Dropout层。通过一个placeholder传入keep_prob比率来控制的。在训练时，我们随机丢弃一部分节点# 的数据来减轻过拟合，预测时则保留全部数据来追求最好的预测性能。keep_prob = tf.placeholder(dtype=tf.float32)h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)# 最后我们将Dropout层的输出连接一个Softmax层，得到最后的概率输出W_fc2 = weight_variable([1024, 10])b_fc2 = bias_variable([10])y_conv = tf.nn.softmax(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)# 定义损失函数为cross entropy和优化器cross_entropy = tf.reduce_mean(-tf.reduce_sum(y_ * tf.log(y_conv), reduction_indices=[1]))train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)# 定义评测准确率的操作correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y_, 1))accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))# 下面开始训练tf.global_variables_initializer().run()for i in range(20000): batch = mnist.train.next_batch(50) if i % 100 == 0: train_accuracy = accuracy.eval(feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0}) print("Step %d, training accuracy %g" % (i, train_accuracy)) train_step.run(feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})print("test accuracy %g" % accuracy.eval(feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})) 
