cnn实现手写体数字识别

数据集 https://download.csdn.net/download/qq_42363032/12737988

import tensorflow as tf
import random
import matplotlib.pyplot as plt
from tensorflow.examples.tutorials.mnist import input_data
import numpy
import tensorflow.contrib as con

tf.set_random_seed(1)

# 获取数据集
mnist = input_data.read_data_sets(r'G:\A_深度学习1\tensorflow\MNIST_data', one_hot=True)

# 定义占位符
X = tf.placeholder('float', [None, 784])    # 处理28*28的图片，这里特征列是784
X_img = tf.reshape(X, shape=[-1, 28, 28, 1])    # 对图片做卷积 要转换成图片形式 28*28的灰度图片
Y = tf.placeholder('float', shape=[None, 10])   # 0-9十个数 是个类别 读数据的时候已经做过one-hot编码了

# 第1层卷积  输入图片数据(?, 28, 28, 1)
'''
卷积核3*3  输入通道1（灰色1 彩色3） 输出通道32（卷积核的个数）
'''
W1 = tf.Variable(tf.random_normal([3, 3, 1, 32]))
'''
tf.nn.conv2d 卷积运算  
X_img 被卷积图片 W1 卷积核
strides=[1, 1, 1, 1]  步长 [1, 垂直方向步长, 水平方向步长, 1]
padding='SAME'  自动填充
'''
L1 = tf.nn.conv2d(X_img, W1, strides=[1, 1, 1, 1], padding='SAME')  # 卷积输出 （?, 28, 28, 32)
L1 = tf.nn.relu(L1)  # relu函数非线性激活
'''
池化操作(压缩数据)  最大值池化
ksize=[1, 2, 2, 1] 池化层尺寸
'''
L1 = tf.nn.max_pool(L1, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')   # 池化输出 (?, 14, 14, 32)

# 第二层卷积  输入图片数据(?, 14, 14, 32)
W2 = tf.Variable(tf.random_normal([3, 3, 32, 64]))  # 卷积核3*3 输入通道32，输出通道64
L2 = tf.nn.conv2d(L1, W2, strides=[1, 1, 1, 1], padding='SAME')  # 卷积输出  (?, 14, 14, 64)
L2 = tf.nn.relu(L2)
L2 = tf.nn.max_pool(L2, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')   # 池化输出 (?, 7, 7, 64)

# 变成一维向量
L2_flat = tf.reshape(L2, shape=[-1, 7 * 7 * 64])  # 变成一维向量 (?, 3136)

# 卷积后的图片进行全连接
W3 = tf.get_variable(name='W3', shape=[7 * 7 * 64, 10], initializer=con.layers.xavier_initializer())    #  Xavier初始化器就是让权重不大不小，刚好合适
b = tf.Variable(tf.random_normal([10]))
logist = tf.matmul(L2_flat, W3) + b

# 代价和Adam自适应优化器
cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=logist, labels=Y))
optimizer = tf.train.AdamOptimizer(learning_rate=0.001).minimize(cost)

# 会话训练
epochs = 15  # 训练总批次
batch_size = 100    # 每批训练的样本数
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    for epoch in range(epochs):
        avg_cost = 0    # 平均代价
        total_batch = int(mnist.train.num_examples / batch_size)    # 每一批次下需要训练的次数
        for i in range(total_batch):
            xs, ys = mnist.train.next_batch(batch_size)     # 每次获得100个数据集
            c, _ = sess.run([cost, optimizer], feed_dict={X: xs, Y: ys})
            avg_cost += c / total_batch     # 计算每个批次的平均代价

        print(f'epoch: {epoch + 1}, {avg_cost}')


    # 测试集检验准确率
    correct_prediction = tf.equal(tf.argmax(logist, 1), tf.argmax(Y, 1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
    print('Accuracy:', sess.run(accuracy, feed_dict={X: mnist.test.images[:5000], Y: mnist.test.labels[:5000]}))

    # 在测试集中随机抽取一个样本进行测试
    r = random.randint(0, mnist.test.num_examples - 1)
    print("Label: ", sess.run(tf.argmax(mnist.test.labels[r:r + 1], 1)))
    print("Predict  ion: ", sess.run(tf.argmax(logist, 1), feed_dict={X: mnist.test.images[r:r + 1]}))

    # 生成一个24位图片（28*28）进行测试
    img = plt.imread(r'G:\A_深度学习1\tensorflow\1.bmp')    # 导入测试图片
    gra = numpy.array([1., 0., 0.])
    greyimg = numpy.dot(255 - img, gra) / 255
    print("Prediction: ", sess.run(tf.argmax(logist, 1), feed_dict={X: greyimg.reshape([1, 784])}))
    plt.imshow(greyimg, cmap='Greys', interpolation='nearest')
    plt.show()