我们采用的卷积神经网络是两层卷积层，两层池化层和两层全连接层
我们使用的数据是mnist数据，数据训练集的数据是50000*28*28*1 因为是黑白照片，所以通道数是1
第一次卷积采用64个filter， 第二次卷积采用128个filter，池化层的大小为2*2，我们采用的是两次全连接
第一步：导入数据
import numpy as np import tensorflow as tf import matplotlib.pyplot as plt from tensorflow.examples.tutorials.mnist import input_data mnist = input_data.read_data_sets('data/', one_hot=True)
第二步： 初始化函数
# 构造初始化参数， 方差为0.1 n_input = 784 n_output = 10 weights = { 'wc1' : tf.Variable(tf.truncated_normal([3, 3, 1, 64], stddev=0.1)), 'wc2' : tf.Variable(tf.truncated_normal([3, 3, 64, 128], stddev=0.1)), 'wd1' : tf.Variable(tf.truncated_normal([7*7*128, 1024], stddev=0.1)), 'wd2' : tf.Variable(tf.truncated_normal([1024, n_output], stddev=0.1)) } biases = { 'b1' : tf.Variable(tf.truncated_normal([64], stddev=0.1)), 'b2' : tf.Variable(tf.truncated_normal([128], stddev=0.1)), 'bd1' : tf.Variable(tf.truncated_normal([1024], stddev=0.1)), 'bd2' : tf.Variable(tf.truncated_normal([n_output], stddev=0.1)) }
第三步： 构造前向传播卷积函数，两次卷积，两次池化，两次全连接
def conv_basic(_input, _w, _b, _keepratio): _input_r = tf.reshape(_input, shape=[-1, 28, 28, 1]) #进行卷积操作 _conv1 = tf.nn.conv2d(_input_r, _w['wc1'], strides=[1, 1, 1, 1], padding='SAME') # 使用激活函数 _conv1 = tf.nn.relu(tf.nn.bias_add(_conv1, _b['bc1'])) # 进行池化操作, padding='SAME', 表示维度不足就补齐 _pool1 = tf.nn.max_pool(_conv1, ksize=[1, 2, 2, 1], stride=[1, 2, 2, 1], padding='SAME') #去除一部分数据 _pool1_dr1 = tf.nn.dropout(_pool1, _keepratio) #第二次卷积操作 _conv2 = tf.nn.conv2d(_pool1_dr1, _w['wc1'], strides=[1, 1, 1, 1], padding='SAME') # 使用激活函数 _conv2 = tf.nn.relu(tf.nn.bias_add(_conv1, _b['bc1'])) # 进行池化操作 _pool2 = tf.nn.max_pool(_conv1, ksize=[1, 2, 2, 1], stride=[1, 2, 2, 1], padding='SAME') _pool_dr2 = tf.nn.dropout(_pool1, _keepratio) # 第一次全连接操作 # 对_pool_dr2 根据wd1重新构造函数 _densel = tf.reshape(_pool_dr2, [-1, _w['wd1'].get_shape().as_list()[0]]) _fcl = tf.nn.relu(tf.add(tf.matmul(_densel, _w['wd1'], _b['bd1']))) _fc_dr1 = tf.nn.dropout(_fcl, _keepratio) # 第二次全连接 _out = tf.add(tf.matmul(_fc_dr1, _w['wd2']), _b['bd2']) out = {'input_r': _input_r, 'conv1': _conv1, 'pool1': _pool1, 'pool1_dr1': _pool_dr1, 'conv2': _conv2, 'pool2': _pool2, 'pool_dr2': _pool_dr2, 'dense1': _dense1, 'fcl': _fcl, 'fc_dr1': _fc_dr1, 'out': _out } return out
第四步： 构造cost函数，和准确值函数
x = tf.placeholder(tf.float32, [None, n_input])
y = tf.placeholder(tf.float32, [None, n_output])
keepratio = tf.placeholder(tf.float32)
# 构造cost函数 #获得预测结果 _pred =conv_basic(x, weights, biases, keepratio)['out'] # 输入预测结果与真实值构造cost 函数 cost = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(_pred, y)) # 优化函数使得cost最小 optm = tf.train.AdamOptimizer(learning_rate=0.001).minimize(cost) # 计算准确率 _corr = tf.equal(tf.argmax(_pred, 1), tf.argmax(y, 1)) accr = tf.reduce_mean(tf.cast(_corr, tf.float32))
第五步： 训练模型，降低cost，提升精度
init = tf.global_variables_initializer() # 进行训练 sess = tf.Session() sess.run(init) #迭代次数 training_epochs = 15 # 每次训练的样本数 batch_size = 16 #循环打印的次数 display_step = 1 for epoch in range(training_epochs): avg_cost = 0. #total_batch = int(mnist.train.num_examples/batch_size) total_batch = 10 # Loop over all batches for i in range(total_batch): # 提取训练数据和标签 batch_xs, batch_ys = mnist.train.next_batch(batch_size) #训练模型优化参数 sess.run(optm, feed_dict={x: batch_xs, y: batch_ys, keepratio:0.7}) # 加和损失值 avg_cost += sess.run(cost, feed_dict={x: batch_xs, y: batch_ys, keepratio:1.})/total_batch # Display logs per epoch step if epoch % display_step == 0: print ("Epoch: %03d/%03d cost: %.9f" % (epoch, training_epochs, avg_cost)) train_acc = sess.run(accr, feed_dict={x: batch_xs, y: batch_ys, keepratio:1.}) print (" Training accuracy: %.3f" % (train_acc)) #test_acc = sess.run(accr, feed_dict={x: testimg, y: testlabel, keepratio:1.}) #print (" Test accuracy: %.3f" % (test_acc)) print ("OPTIMIZATION FINISHED")