卷积神经网络实现手写数字识别---My way of AI 22

在手写数字识别这篇文章中我们实现了用全连接神经网络实现图像识别。这篇文章我们在卷积神经网络的理论基础之上实现手写数字识别。

流程

1.定义卷积模型

卷积模型是我们自定义的，我们用来两侧卷积激活池化和一层全连接层实现卷积模型。

 #卷积操作
        con=tf.nn.conv2d(x_reshape,weight,strides=[1,1,1,1],padding="SAME")

        #加上偏置
        con=con+biases

        #激活
        actived_conv=tf.nn.relu(con)

        #池化

        pool = tf.nn.max_pool(actived_conv, ksize=[1, 2,2, 1], strides=[1, 2, 2, 1], padding="SAME")

2.主函数

主函数部分我们还是用交叉熵衡量损失，用梯度下降去优化。

 with tf.variable_scope("soft_max"):
        # 求平均交叉熵损失
        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))

    # 梯度下降求出损失
    with tf.variable_scope("optimaizer"):
        train_op = tf.train.GradientDescentOptimizer(0.001).minimize(loss)

3.开启会话

开启绘画后利用op的依赖性去传递参数。去喂数据

sess.run(train_op, feed_dict={x: mnist_x, y_true: mnist_y})

代码

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

def weight_variable(shape):
    """
    初始化权重的函数
    :shape:形状
    :return: w
    """
    w=tf.Variable(tf.random_normal(shape=shape,mean=0.0,stddev=1.0))
    return w


def bias_varible(shape):
    """
    定义初始化偏置函数
    :param shape: 形状
    :return: b
    """
    bias=tf.Variable(tf.constant(0.0,shape=shape))
    return bias

def model():
    """
    自定义卷积模型
    :return:
    """
    #准备占位符
    with tf.variable_scope("data"):
        x=tf.placeholder(tf.float32,[None,784])
        y_true=tf.placeholder(tf.int32,[None,10])

    #卷积层1,卷积激活池化
    with tf.variable_scope("con1"):
        #对x形状改变
        x_reshape=tf.reshape(x,[-1,28,28,1])

        shape_weight=[5,5,1,32]
        shape_bias=[32]
        weight=weight_variable(shape_weight)

        biases=bias_varible(shape_bias)

        #卷积操作
        con=tf.nn.conv2d(x_reshape,weight,strides=[1,1,1,1],padding="SAME")

        #加上偏置
        con=con+biases

        #激活
        actived_conv=tf.nn.relu(con)

        #池化

        pool = tf.nn.max_pool(actived_conv, ksize=[1, 2,2, 1], strides=[1, 2, 2, 1], padding="SAME")

    #卷积层2
    with tf.variable_scope("con2"):
        shape_weight=[5,5,32,64]
        biases=bias_varible([64])
        weight=weight_variable(shape_weight)

        #卷积，激活，池化
        pool=tf.nn.max_pool(tf.nn.relu(tf.nn.conv2d(pool,weight,strides=[1,1,1,1],padding="SAME")+biases),
                            ksize=[1, 2,2, 1], strides=[1, 2, 2, 1], padding="SAME")


    #全连接层
    with tf.variable_scope("full_con"):
        w_fc=weight_variable([7*7*64,10])
        b_fc=bias_varible([10])


        x_fc_reshape=tf.reshape(pool,[-1,7*7*64])
        #进行矩阵运算，得出每个样本计算结果
        y_predict=tf.matmul(x_fc_reshape,w_fc)+b_fc

    return x,y_true,y_predict



def conv2():
    """
    卷积实现手写数字识别
    :return: None
    """
    mnist = input_data.read_data_sets("./data", one_hot=True)
    #定义模型得出输出,第一步的向前传播
    x, y_true, y_predict=model()


    with tf.variable_scope("soft_max"):
        # 求平均交叉熵损失
        loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict))

    # 梯度下降求出损失
    with tf.variable_scope("optimaizer"):
        train_op = tf.train.GradientDescentOptimizer(0.001).minimize(loss)

    # 计算准确率
    with tf.variable_scope("acc"):
        equal_list = tf.equal(tf.argmax(y_true, 1), tf.argmax(y_predict, 1))

        accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))

    #开启绘画运行
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())

        for i in range(1000):
            # 取出真实存在的特征值和目标值
            mnist_x, mnist_y = mnist.train.next_batch(50)

            # 运行训练
            sess.run(train_op, feed_dict={x: mnist_x, y_true: mnist_y})

            # 写入每步训练的值
            # summary = sess.run(merged, feed_dict={x: mnist_x, y_true: mnist_y})
            # filewrite.add_summary(summary, i)

            print("训练第%d步，准确率为%f" % (i, sess.run(accuracy, feed_dict={x: mnist_x, y_true: mnist_y})))

if __name__ == '__main__':
    conv2()

author ：specyue@mail.ustc.edu.cn 欢迎交流

github：https://github.com/zhangyuespec/AI