写在前面的话
文章总结自北京大学曹健老师的mooc,点击查看课程,通过学习,我自己撸了一遍代码,现在放在下面。并写了一些自己的认识
环境版本
tensorflow==1.3.0
python==3.5.6
操作系统为ubuntu16
手写数字识别的思路
1、首先,我们用tensorflow官方给出的mnist数据集来训练出一个模型
2、其次,我们用tensorflow官方给出的mnist数据集来测试这个模型的精度
3、将我们自己手写的数字图片喂入神经网络,得出预测结果
代码思路
1、我们先设计一个前向传播网络,这个py文件中定义神经络的输入、参数和输出,定义前向传播过程
get_weight为神经元w参数生成函数
get_bias为偏置b生成函数
forward复现了网络的结构
#0导入模块
import tensorflow as tf
INPUT_NODE = 784
OUTPUT_NODE = 10
LAYER1_NODE = 500
# #定义神经络的输入、参数和输出,定义前向传播过程
def get_weight(shape, regularizer):
w = tf.Variable(tf.truncated_normal(shape, stddev=0.1))#截断的产生正态分布的随机数,即随机数与均值的差值若大于两倍的标准差,则重新生成。stddev:标准差
if regularizer != None: tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))#正则化给每个w加权重
return w
def get_bias(shape):
b = tf.Variable(tf.zeros(shape))
return b
def forward(x, regularizer):
w1 = get_weight([INPUT_NODE, LAYER1_NODE], regularizer)
b1 = get_bias([LAYER1_NODE])
y1 = tf.nn.relu(tf.matmul(x, w1) + b1)#非线性函数relu的输出
w2 = get_weight([LAYER1_NODE, OUTPUT_NODE], regularizer)
b2 = get_bias([OUTPUT_NODE])
y = tf.matmul(y1, w2) + b2 #输出层不过激活函数,因为要保证概率分布的均匀?
return y
2、在forward.py文件中,我们定义了神经网络的一些个参数,解释如代码注释,其中我们定义了代码的反向传播纠正参数的结构。定义了ckpt来实现断点续训,并在控制台输出了训练的情况即损失 值的大小。
#0导入模块,生成模拟数据集
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import os
BATCH_SIZE = 200#每轮训练喂入数据的量
LEARNING_RATE_BASE = 0.1#学习率
LEARNING_RATE_DECAY = 0.99#指数衰减率
REGULARIZE = 0.0001#正则化权重
STEPS = 50000
MOVING_AVERAGE_DECAY = 0.99#衰减因子
MODEL_SAVE_PATH = "./model/"
MODEL_NAME = "mnist_model"
def backward(mnist):
x = tf.placeholder(tf.float32, [None, mnist_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnist_forward.OUTPUT_NODE])
y = mnist_forward.forward(x, REGULARIZE)#复现网络结构,推算出预测值
global_step = tf.Variable(0, trainable=False)
#定义损失函数,引入了正则化的损失函数,并和交叉熵一起使用
ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits=y, labels=tf.argmax(y_, 1))#计算logits和labels之间的稀疏softmax交叉熵
cem = tf.reduce_mean(ce)
loss = cem + tf.add_n(tf.get_collection('losses'))
#定义指数衰减学习率
learning_rate = tf.train.exponential_decay(
LEARNING_RATE_BASE,
global_step,
mnist.train.num_examples/BATCH_SIZE,
LEARNING_RATE_DECAY,
staircase=True)#如果staircase=True,那就表明每decay_steps次计算学习速率变化,更新原始学习速率,如果是False,那就是每一步都更新学习速率
#定义反向传播方法:包涵正则化
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step=global_step)
#滑动平均值
ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
ema_op = ema.apply(tf.trainable_variables())
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name='train')
saver = tf.train.Saver()
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
ckpt = tf.train.get_checkpoint_state(MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
for i in range(STEPS):
xs, ys = mnist.train.next_batch(BATCH_SIZE)
_, loss_value, step = sess.run([train_op, loss, global_step], feed_dict={x: xs,y_: ys})
if i % 1000 == 0:
print("After %d training step(s), loss on training batch is %g."%(step, loss_value))
saver.save(sess, os.path.join(MODEL_SAVE_PATH, MODEL_NAME), global_step=global_step)
def main():
mnist = input_data.read_data_sets("./data/", one_hot=True)
backward(mnist)
if __name__ == '__main__':
main()
3、以上神经网络的结构就算实现了,我们既然已经得到了这个模型,就肯定想知道这个模型是否可以解决我们现在要面临的问题,我们编写一个测试代码来检查神经网络的模型精度。测试集代码也是tensorflow官方给出的mnist数据集。我们复现这个网络,并测试效果,基本上一轮之后你就能观察到精确率达到了90%+。
import time
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import mnist_forward
import mnist_backward
TEST_INTNERVAL_SECS = 10
def test(mnist):
with tf.Graph().as_default() as g:
x = tf.placeholder(tf.float32, [None, mnist_forward.INPUT_NODE])
y_ = tf.placeholder(tf.float32, [None, mnist_forward.OUTPUT_NODE])
y = mnist_forward.forward(x, None)
ema = tf.train.ExponentialMovingAverage(mnist_backward.MOVING_AVERAGE_DECAY)
ema_restore = ema.variables_to_restore()
saver = tf.train.Saver(ema_restore)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
while True:
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state(mnist_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
global_step = ckpt.model_checkpoint_path.split('/')[-1].split('-')[-1]
accuracy_score = sess.run(accuracy, feed_dict={x: mnist.test.images, y_:mnist.test.labels})
print("After %s training step(s), test accuracy = %g"%(global_step, accuracy_score))
else:
print('No checkpoint file found')
return
time.sleep(TEST_INTNERVAL_SECS)
def main():
mnist = input_data.read_data_sets("./data/", one_hot=True)
test(mnist)
if __name__ == '__main__':
main()
4、最后,自己愉快地拍下一张照片,来看看这个亲儿子模型是不是好使。粘一下代码:
import tensorflow as tf
import numpy as np
import mnist_forward
import mnist_backward
from PIL import Image
def restore_model(testPicArr):
#创建一个默认图,在该图中执行以下操作(多数操作和train中一样)
with tf.Graph().as_default() as tg:
x = tf.placeholder(tf.float32, [None, mnist_forward.INPUT_NODE])
y = mnist_forward.forward(x, None)
preValue = tf.argmax(y, 1)#得到概率最大的预测值
variable_averages = tf.train.ExponentialMovingAverage(mnist_backward.MOVING_AVERAGE_DECAY)
variable_to_restore = variable_averages.variables_to_restore()
saver = tf.train.Saver(variable_to_restore)
with tf.Session() as sess:
#chevkpoint 文件定位到最新保存的模型
ckpt = tf.train.get_checkpoint_state(mnist_backward.MODEL_SAVE_PATH)
if ckpt and ckpt.model_checkpoint_path:
saver.restore(sess, ckpt.model_checkpoint_path)
preValue = sess.run(preValue, feed_dict={x:testPicArr})
return preValue
else:
print("No checkpoint file found'")
return -1
def pre_pic(picName):
img = Image.open(picName)
reIm = img.resize((28,28), Image.ANTIALIAS)
im_arr = np.array(reIm.convert('L'))
threshold = 180#设定合理的阈值
for i in range(28):
for j in range(28):
im_arr[i][j] = 255 - im_arr[i][j]
if(im_arr[i][j] < threshold):
im_arr[i][j] = 0
else:
im_arr[i][j] = 255
nm_arr = im_arr.reshape([1, 784])
nm_arr = nm_arr.astype(np.float32)
img_ready = np.multiply(nm_arr, 1.0/255.0)
return img_ready
def application():
# testNum = int(input("input the number of test pictures:"))
testNum = 1
for i in range(testNum):
# testPic = input("the path of test pictures:")
testPic = 'pic/1.jpg'
testPicArr = pre_pic(testPic)
preValue = restore_model(testPicArr)
print("The prediction number is : ", preValue)
if __name__ == '__main__':
application()
5、图片如下:
识别结果如下:
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist# python3 hand_write_app.py
The prediction number is : [2]
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist# cd pic
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist/pic# ls
1.jpg
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist/pic# cd ./
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist/pic# cd ../
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist# python3 hand_write_app.py
The prediction number is : [7]
root@iZ2zef0icee95uw35ttpgmZ:/home/kai/mnist# python3 hand_write_app.py
The prediction number is : [5]
6、分析:
大家先只看[ ]里面的输出结果,因为中间我使用pscp拷贝替换了图片
识别了三张图片,最后一张6识别成了5,其他两张识别正确,思考和心得体会见下文
---------- 代码已经结束了,就是上面的四个py文件,接下来你看到的是我的分析过程 ------
心得体会
分析一:这里我使用了三张图片,一开始都不能识别,后来我经过调试,发现是阈值 threshold 给的太低了,我一开始给的是50,然后一张也识别不了,代码是没有错的,应该就是在图像预处理这方面出了差错,我先打印输出了 1图片的灰度矩阵 和 2当阈值为50时的 图片灰度反转后的值 和 3当阈值为50时的 图片的样子,测试结果如下:
1图片的灰度矩阵
2当阈值为50时的 图片灰度反转后的值 :明显看出阈值选择错误导致2的特征失真
3当阈值为50时的 图片的样子:
自己脑补一下,反正全是黑的不能训练
分析二:
当阈值为180,我自己调的参数结果明显变好,全是明显的轮廓,所以可以用于训练。
结论
图像的预处理真的很重要,我们要保证图片的特征存在。