【TensorFlow实战Google深度学习框架】学习笔记（图像识别与卷积神经网络）

1、图像识别经典数据集Cifar

cifar是32x32x3的彩色图片集

2、卷积神经网络简介

3、卷积神经网络常用的结构

3.1卷积层

Filter的深度就是Filter的个数

3.2池化层

4、经典卷积神经网络模型

卷积神经网络结构的问题

卷积层和池化层的参数配置问题

5、卷积网络的迁移学习

迁移学习可以总结为在别人成功的模型（网络结构和训练好的参数）的基础上修改、增减自己的网络结构。这样做的好处是缩短模型搭建的时间，减少参数的训练量。

附录

conv_mnist_infernece.py

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

"""
使用LeNet-5模型实现卷积神经网络的前向传播
"""
import tensorflow as tf

INPUT_NODE = 784
OUTPUT_NODE = 10

IMAGE_SIZE = 28
NUM_CHANNELS = 1
NUM_LABELS = 10

# 第一层卷积层的尺寸和深度
CONV1_DEEP = 32
CONV1_SIZE = 5

# 第二层卷积层的尺寸和深度
CONV2_DEEP = 64
CONV2_SIZE = 5

# 全连接层的节点个数
FC_SIZE = 512

# 定义卷积神经网络的前向传播过程。 注意区别训练过程和测试过程。
def infernece(input_tensor, train, regularizer):

    # 声明第一层卷积层的变量，并实现前向传播过程。
    # 定义的卷积层输入为 28*28*1 的原始MNIST图片像素， 使用了全0填充，所以输出为 28*28*32
    with tf.variable_scope("layer1-conv1"):
        conv1_weights = tf.get_variable(
            "weight", [CONV1_SIZE, CONV1_SIZE, NUM_CHANNELS, CONV1_DEEP],
            initializer=tf.truncated_normal_initializer(stddev=0.1))
        conv1_biases = tf.get_variable(
            "bias", [CONV1_DEEP], initializer=tf.constant_initializer(0.0))

        # 使用边长为5 深度为32的过滤器，移动步长1 且使用全填充
        conv1 = tf.nn.conv2d(
            input_tensor, conv1_weights, strides=[1, 1, 1, 1], padding="SAME")
        relu1 = tf.nn.relu(tf.nn.bias_add(conv1, conv1_biases))

    # 实现第二层池化层的前向传播，使用了最大池化层。过滤器边长为2，使用全0填充，步长为2
    # 这一层的输入为上一层的输出 即 28*28*32 ，输出这为 14*14*32 池化不会改变深度
    with tf.name_scope("layer2-pooll"):
        pool1 = tf.nn.max_pool(
            relu1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")

    # 声明第三层卷积层的变量，并实现前向传播。
    # 这一层的输入为  14*14*32  输出为  14*14*64
    with tf.variable_scope("layer3-conv2"):
        conv2_weights = tf.get_variable(
            "weight", [CONV2_SIZE, CONV2_SIZE, CONV1_DEEP, CONV2_DEEP],
            initializer=tf.truncated_normal_initializer(stddev=0.1))
        conv2_biases = tf.get_variable(
            "bias", [CONV2_DEEP], initializer=tf.constant_initializer(0.0))

        # 使用边长为5， 深度为64的过滤器，步长为1，使用全0填充
        conv2 = tf.nn.conv2d(pool1, conv2_weights, strides=[1, 1, 1, 1], padding="SAME")
        relu2 = tf.nn.relu(tf.nn.bias_add(conv2, conv2_biases))

    # 实现第四层池化层的前向传播过程。和第二层类似，这层的输入为 14*14*64 输出为 7*7*64
    with tf.name_scope("layer4-pool2"):
        pool2 = tf.nn.max_pool(
            relu2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding="SAME")

    # 将第四层池化层的输出转化为第五层全连接层的输入格式。第四层输出 7*7*64
    # 这里需要将 7*7*64 拉成一个向量。
    pool_shape = pool2.get_shape().as_list()

    # 拉成一个向量，长度为之前矩阵的长宽及深度的乘积 pool_shape[0]为一个batch中数据个数  这里可以理解吧
    nodes = pool_shape[1] * pool_shape[2] * pool_shape[3]

    # 通过tf.reshape 函数将第四层的输出变成一个batch的向量
    reshaped = tf.reshape(pool2, [pool_shape[0], nodes])

    # 声明第五层全连接层的变量，并实现前向传播过程。输入的长度为7*7*64 = 3136 输出是512长度
    with tf.variable_scope("layer5-fc1"):
        fc1_weights = tf.get_variable(
            "weight", [nodes, FC_SIZE], initializer=tf.truncated_normal_initializer(stddev=0.1))
        # 只有全连接层的权重需要加入正则化
        if regularizer != None:
            tf.add_to_collection("losses", regularizer(fc1_weights))
        fc1_biases = tf.get_variable(
            "bias", [FC_SIZE], initializer=tf.constant_initializer(0.1))
        fc1 = tf.nn.relu(tf.matmul(reshaped, fc1_weights) + fc1_biases)

        # 使用dropout在训练时随机将部分节点的输出改为0，避免过拟合。一般用在全连接层
        if train:
            fc1 = tf.nn.dropout(fc1, 0.5)

    # 声明第六层全连接的变量和前向传播过程。这层长度为512，输出为长度10.通过Softmax分类得到结果
    with tf.variable_scope("layer6-fc2"):
        fc2_weights = tf.get_variable(
            "weight", [FC_SIZE, NUM_LABELS], initializer=tf.truncated_normal_initializer(0.1))
        if regularizer != None:
            tf.add_to_collection("losses", regularizer(fc2_weights))
        fc2_biases = tf.get_variable(
            "bias", [NUM_LABELS], initializer=tf.constant_initializer(0.1))
        # 这里不需要激活函数
        logit = tf.matmul(fc1, fc2_weights) + fc2_biases

    return logit

conv_mnist_train.py

import tensorflow as tf
import conv_mnist_infernece
from tensorflow.examples.tutorials.mnist import input_data
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

# 配置神经网络参数
BATCH_SIZE  = 100  # 数据模型
LEARNINT_RATE_BASE = 0.8  # 学习率
LEARNINT_RATE_DECAY = 0.99  # 学习衰减率
REGULARAZTION_RATE = 0.0001  # 正则化损失
TRAINING_STEPS = 30000  # 训练步数
MOVING_AVERAGE_DECAY = 0.99  # 滑动平均衰减率

# 保存模型路径
MODEL_SAVE_PATH = "./conv/"
MODEL_NAME = "model.ckpt"

def train(mnist):
    # 调整输入数据的格式，输入为一个四维矩阵
    x = tf.placeholder(dtype=tf.float32,
                       shape=[BATCH_SIZE,
                              conv_mnist_infernece.IMAGE_SIZE,
                              conv_mnist_infernece.IMAGE_SIZE,
                              conv_mnist_infernece.NUM_CHANNELS],
                       name="x-input")
    y_ = tf.placeholder(tf.float32, [None, conv_mnist_infernece.OUTPUT_NODE], name='y-out')

    # 计算L2正则化损失项
    regularizer = tf.contrib.layers.l2_regularizer(REGULARAZTION_RATE)
    y = conv_mnist_infernece.infernece(x, False, regularizer)
    global_step = tf.Variable(0, trainable=False)

    # 定义损失函数、学习率、滑动平均操作 及训练过程
    variable_averages = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
    variable_averages_op = variable_averages.apply(tf.trainable_variables())
    cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))
    cross_entropy_mean = tf.reduce_mean(cross_entropy)
    loss = cross_entropy_mean + tf.add_n(tf.get_collection('losses'))

    # 指数衰减
    learning_rate = tf.train.exponential_decay(
        LEARNINT_RATE_BASE,
        global_step,
        mnist.train.num_examples / BATCH_SIZE,
        LEARNINT_RATE_DECAY)
    # 在训练神经网络模型时，每过一遍数据既要通过反向传播来更新神经网络中的参数
    train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)
    # 同时也要更新每一个参数的滑动平均值。tf.group 和 tf.control_dependencies 两种方法
    # train_op = tf.group(train_step, variavle_averages_op, name='train')
    with tf.control_dependencies([train_step, variable_averages_op]):
        train_op = tf.no_op(name='train')

    saver = tf.train.Saver()
    with tf.Session() as sess:
        tf.global_variables_initializer().run()
        # 开始训练过程
        for i in range(TRAINING_STEPS):
            xs, ys = mnist.train.next_batch(BATCH_SIZE)
            # print(xs.shape)
            # print(ys.shape)
            reshape_xs = tf.reshape(xs, [BATCH_SIZE,
                                         conv_mnist_infernece.IMAGE_SIZE,
                                         conv_mnist_infernece.IMAGE_SIZE,
                                         conv_mnist_infernece.NUM_CHANNELS])
            reshape_xs = reshape_xs.eval()
            _, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: reshape_xs, y_: ys})
            if i % 10 == 0:
                print("After %d training step(s), loss on training average batch is %g ." % (step, loss_value))
                # 保存模型
                saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step)


def main(argv=None):
    mnist = input_data.read_data_sets("./MNIST", one_hot=True)
    train(mnist)


if __name__ == "__main__":
    tf.app.run()