[人工智能]去噪自编码器tensorflow实现

import tensorflow.contrib.layers as lays
import numpy as np
from skimage import transform
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import matplotlib.pyplot as plt

#定义具有编码器和解码器的网络
def autoencoder(inputs):
    net = lays.conv2d(inputs,32,[5,5],stride=2,padding='SAME')
    net = lays.conv2d(net,16,[5,5],stride=2,padding='SAME')
    net = lays.conv2d(net,8,[5,5],stride=4,padding='SAME')
    # 解码器
    net = lays.conv2d_transpose(net,16,[5,5],stride=4,padding='SAME')
    net = lays.conv2d_transpose(net,32,[5,5],stride=2,padding='SAME')
    net = lays.conv2d_transpose(net,1,[5,5],stride=2,padding='SAME',activation_fn=tf.nn.tanh)
    return net

def resize_batch(imgs):
    #该函数将图像调整为32×32，这样维数可以减少一半
    imgs = imgs.reshape((-1,28,28,1))
    resized_imgs =np.zeros((imgs.shape[0],32,32,1))
    for i in range(imgs.shape[0]):
        resized_imgs[i,...,0] = transform.resize(imgs[i, ..., 0],(32,32))
    return resized_imgs

##该函数引入高斯噪声
def noisy(image):
    row, col = image.shape
    mean = 0
    var = 0.1
    sigma = var**0.5
    gauss = np.random.normal(mean,sigma,(row,col))
    gauss = gauss.reshape(row,col)
    noisy = image + gauss
    return noisy

##椒盐噪声先不写

#    定义操作
#与重建信号进行比较的输入
a_e_inputs =tf.placeholder(tf.float32,(None,32,32,1))
#输入网络（MINST图像）
a_e_inputs_noise = tf.placeholder(tf.float32,(None,32,32,1))
a_e_outputs  = autoencoder(a_e_inputs_noise)#创建自编码器网络

loss = tf.reduce_mean(tf.square(a_e_outputs - a_e_inputs))# 计算均方误差

train_op = tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)
#初始化网络
init = tf.global_variables_initializer()

batch_size = 1000  #每个批量的样本数
epoch_num = 200    #网络训练的迭代次数
lr = 0.001         #学习率

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

batch_per_ep = mnist.train.num_examples//batch_size

with tf.Session() as sess:
    sess.run(init)
    for epoch in range(epoch_num):
        batch_img, batch_label=mnist.train.next_batch(batch_size)

        batch_img = batch_img.reshape((-1,28,28,1))
        batch_img = resize_batch(batch_img)

        image_arr = []
        for i in range(len(batch_img)):
            img = batch_img[i, :, :, 0]
            img = noisy(img)
            image_arr.append(img)
        image_arr = np.array(image_arr)
        image_arr = image_arr.reshape(-1,32,32,1)
        _, c = sess.run([train_op, loss], feed_dict={a_e_inputs_noise: image_arr, a_e_inputs: batch_img})
        print('Epoch:{} - cost = {:.5f}'.format((epoch + 1),c))

#测试训练网络
    batch_img, batch_label = mnist.test.next_batch(50)
    batch_img = resize_batch(batch_img)
    image_arr = []
    for i in range(50):
        img = batch_img[i,:,:,0]
        img = noisy(img)
        image_arr.append(img)
    image_arr = np.array(image_arr)
    image_arr = image_arr.reshape(-1,32,32,1)

    reconst_img = sess.run([a_e_outputs],feed_dict={a_e_inputs_noise:image_arr})[0]
    #画出重构的图像
    plt.figure(1)
    plt.title('Input Noisy Images')
    for i  in range(50):
        plt.subplot(5,10,i+1)
        plt.imshow(image_arr[i, ...,0],cmap ='gray')
    plt.figure(2)
    plt.title('Re-constructed Images')
    for i  in range(50):
        plt.subplot(5,10,i+1)
        plt.imshow(reconst_img[i, ...,0],cmap ='gray')
    plt.show()