正文
使用CNN做数字识别和人脸识别
提示：扫一扫查出行【扫一扫了解最新限行尾号】
复制提示
- 上次写的一层神经网络也都贴这里了。
- 我有点困，我先睡觉，完了我再修改
- 这个代码写法不太符合工业代码的规范，仅仅是用来学习的的。还望各位见谅
import sys,os
sys.path.append(os.pardir)
import numpy as np
from tensorflow.examples.tutorials.mnist import input_data
from PIL import Image
import tensorflow as tf class mmmmser:
id=""
name="" def __int__(self,name,id):
self.name = name
self.id = id
def setself(self,name,id):
self.name = name
self.id = id
def setname(self,name):
self.name = name
def getname(self):
return self.name
def setid(self, id):
self.id = id
def getid(self):
return self.id #交叉熵损失函数
#值越小对应得越是预测的值
#预测的值越大，反而不是正确的预测
#减少了其他结果的影响
def cross_entropy_error(y, t):
delta = 1e-7
return -np.sum(t * np.log(y + delta))
#均方差损失函数
#均方误差，对于的结果相减，平方求和。
#有其他结果的影响。
def mean_squared_error(y,t):
return 0.5*np.sum((y-t)**2)
#图像显示函数
def showImg():
batch_xs, batch_ys = mnist.train.next_batch (100)
x=np.random.choice(batch_xs.shape[0],4)
myx=batch_xs[0]
myx=myx*255
k=myx.reshape(28,28)
print(myx)
img=Image.fromarray(np.uint8(k))
print(img)
img.show()
def showImgTest(one_img):
myx=one_img*255
k=myx.reshape(28,28)
img=Image.fromarray(np.uint8(k))
img.show()
#获取随机的batch
def GetBatch(train_x,train_y):
train_size = len(train_x)
batch_size = 10 #每次训练的batch数量
batch_mask = np.random.choice (train_size, batch_size)
# print (batch_mask)
x_batch=[]
t_batch=[]
for i in range(len(batch_mask)):
x_batch.append(np.array(train_x[batch_mask[i]]).reshape((-1,784)))
t_batch.append(np.array(train_y[batch_mask[i]]).reshape((-1,10)))
return np.array(x_batch).reshape((10,-1)),np.array(t_batch).reshape((10,-1)) #批量交叉熵
def cross_entropy_error_batch(y, t):
batch_size = y.shape[0]
return -np.sum(t * np.log(y + 1e-7)) / batch_size
def eachfile(filepath,arr,trr,k,a):
print("loading face file..")
pathdir=os.listdir(filepath)
for allpath in pathdir:
if(a>10 and k<1):
break
if k>=0:
child=os.path.join('%s\%s'%(filepath,allpath))
if os.path.isfile(child):
readfile(child,arr,trr,a)
else:eachfile(child,arr,trr,a)
def getpeoplename(num):
user1=mmmmser()
user1.setself("tjl","9")
user2=mmmmser()
user2.setself("lw","8")
user3=mmmmser()
user3.setself("lt","0")
user=[]
user.append(user1)
user.append(user2)
user.append(user3)
k=0
for i in range(len(user)):
if user[i].getid()==num:
k=i
break
else:
continue
return user[k].getname() def readfile(child,arr,trr,a):
a+=1
print(child)
img1=Image.open(child)
limg=img1.convert('L')
k=limg.resize((28,28))
if len(arr)==0:
img = Image.fromarray (np.uint8(k))
img.show()
# print(np.array(k).shape)
arr.append(k)
trr.append([0,0,0,0,0,0,0,0,0,1]) # fopen=open(filename,'r')
# fileread=fopen.read()
# fopen.close()
def trainfaceoftian(arr,trr):
print("train face of tianjingle...")
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
trainImg=tf.reshape(x,[-1,28,28,1])
#第一层网络
w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding='SAME') + b1)
p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #第二层网络
w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding='SAME') + b2)
p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #全连接层
w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3") #cnn输出变形
fp2=tf.reshape(p2,[-1,7*7*64])
fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3) #转为【10】
w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
y_conv = tf.matmul(fp3, w4) + b4 #定义交叉熵
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_,logits=y_conv))
#定义训练
train_step = tf.train.AdamOptimizer(1e-4).minimize (cross_entropy)
# 使用Dropout，keep_prob是一个占位符，训练时为0.5，测试时为1
keep_prob = tf.placeholder(tf.float32)
# 定义测试的准确率
correct_prediction = tf.equal(tf.argmax(y_conv, 1),tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) # 创建Session和变量初始化
sess = tf.InteractiveSession()
sess.run (tf.global_variables_initializer()) #准确标志
maxaccuracy=0 #模型存储
saver = tf.train.Saver(max_to_keep=1)
# 训练1000步
for i in range (1000):
# batch = mnist.train.next_batch(50)
batch_xs, batch_ys = GetBatch (arr, trr)
if i % 100 == 0:
train_accuracy = accuracy.eval (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 1.0})
print ("step %d, training accuracy %g" % (i, train_accuracy))
train_step.run (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 0.5})
if maxaccuracy<train_accuracy:
maxaccuracy=train_accuracy
saver.save (sess, 'tface/tjlmodel.ckpt')
print("Cnn model training with tjl end..") def trainface():
print("start train face...")
filename="C:\\tianjingletianjingle"
arr=[]
trr=[]
eachfile(filename,arr,trr,1,0)
print("length of arr")
print(len(arr))
trainfaceoftian(arr,trr) def loadface():
filename="C:\\tianjingletianjingle"
arr=[]
trr=[]
eachfile(filename,arr,trr,0,0)
return GetBatch(arr,trr) def predicttjl():
print("start predict tjl...with cnn")
meta_path = 'tface/tjlmodel.ckpt.meta' #模型的结构
model_path ='tface/tjlmodel.ckpt' #模型的数据
sess = tf.InteractiveSession()
saver = tf.train.import_meta_graph (meta_path)
saver.restore (sess, model_path)
x = tf.placeholder (tf.float32, [None, 784])
trainImg=tf.reshape(x,[-1,28,28,1])
graph = tf.get_default_graph ()
w1 = graph.get_tensor_by_name ("w1:0")
b1 = graph.get_tensor_by_name ("b1:0")
w2 = graph.get_tensor_by_name ("w2:0")
b2 = graph.get_tensor_by_name ("b2:0")
w3 = graph.get_tensor_by_name ("w3:0")
b3 = graph.get_tensor_by_name ("b3:0")
w4 = graph.get_tensor_by_name ("w4:0")
b4 = graph.get_tensor_by_name ("b4:0") #第一层网络
# w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
# b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding='SAME') + b1)
p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME')
#第二层网络
# w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
# b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding='SAME') + b2)
p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #全连接层
# w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
# b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3") #cnn输出变形
fp2=tf.reshape(p2,[-1,7*7*64])
fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3) #转为【10】
# w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
# b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
y_conv = tf.matmul(fp3, w4) + b4
keep_prob = tf.placeholder (tf.float32) one_img,one_title=loadface()
# img1=tf.reshape(one_img[0],[-1,28,28,1])
# print(img1.shape)
img1=one_img[0].reshape((-1,784))
# print(img1.reshape((0,784)).shape)
temp=sess.run(y_conv,feed_dict={x: img1, keep_prob: 1.0})
a=tf.arg_max(temp,1)
b=tf.arg_max(one_title[0].reshape((-1,10)),1)
if a.eval()==b.eval():
print ('success! his id is :%d his name is:%s'%(b.eval()[0],getpeoplename(b.eval()[0])))
showImgTest(one_img[0])
else:
print("error..") def predictwithCnn():
print("predict with 3 layer cnn..")
meta_path = 'cnn/model.ckpt.meta' #模型的结构
model_path = 'cnn/model.ckpt' #模型的数据
sess = tf.InteractiveSession ()
saver = tf.train.import_meta_graph (meta_path)
saver.restore (sess, model_path) x = tf.placeholder (tf.float32, [None, 784])
trainImg=tf.reshape(x,[-1,28,28,1])
graph = tf.get_default_graph ()
w1 = graph.get_tensor_by_name ("w1:0")
b1 = graph.get_tensor_by_name ("b1:0")
w2 = graph.get_tensor_by_name ("w2:0")
b2 = graph.get_tensor_by_name ("b2:0")
w3 = graph.get_tensor_by_name ("w3:0")
b3 = graph.get_tensor_by_name ("b3:0")
w4 = graph.get_tensor_by_name ("w4:0")
b4 = graph.get_tensor_by_name ("b4:0") #第一层网络
# w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
# b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding='SAME') + b1)
p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME')
#第二层网络
# w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
# b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding='SAME') + b2)
p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #全连接层
# w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
# b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3") #cnn输出变形
fp2=tf.reshape(p2,[-1,7*7*64])
fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3) #转为【10】
# w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
# b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
y_conv = tf.matmul(fp3, w4) + b4
keep_prob = tf.placeholder (tf.float32)
one_img,one_title=mnist.train.next_batch (100)
# img1=tf.reshape(one_img[0],[-1,28,28,1])
# print(img1.shape)
print(one_img.shape)
img1=one_img[0].reshape((-1,784))
print(img1.shape)
temp=sess.run(y_conv,feed_dict={x: img1, keep_prob: 1.0})
a=tf.arg_max(temp,1)
b=tf.arg_max(one_title[0].reshape((-1,10)),1)
print(a.eval())
print(b.eval())
if a.eval()==b.eval():
print ("success! the num is :", (b.eval()[0]))
showImgTest(img1)
else:
print("error..") def CNNmodel():
print("train with 3 layer cnn...")
x = tf.placeholder(tf.float32, [None, 784])
y_ = tf.placeholder(tf.float32, [None, 10])
trainImg=tf.reshape(x,[-1,28,28,1])
#第一层网络
w1=tf.Variable(tf.truncated_normal([5,5,1,32], stddev=0.1),name="w1")
b1=tf.Variable(tf.constant(0.1, shape=[32]),name="b1")
h1=tf.nn.relu (tf.nn.conv2d(trainImg, w1, strides=[1, 1, 1, 1], padding='SAME') + b1)
p1=tf.nn.max_pool(h1, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #第二层网络
w2=tf.Variable(tf.truncated_normal([5,5,32,64], stddev=0.1),name="w2")
b2=tf.Variable(tf.constant(0.1, shape=[64]),name="b2")
h2=tf.nn.relu (tf.nn.conv2d(p1, w2, strides=[1, 1, 1, 1], padding='SAME') + b2)
p2=tf.nn.max_pool(h2, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1], padding='SAME') #全连接层
w3=tf.Variable(tf.truncated_normal([7*7*64,1024], stddev=0.1),name="w3")
b3=tf.Variable(tf.constant(0.1, shape=[1024]),name="b3") #cnn输出变形
fp2=tf.reshape(p2,[-1,7*7*64])
fp3=tf.nn.relu(tf.matmul(fp2, w3) + b3) #转为【10】
w4 = tf.Variable(tf.truncated_normal([1024, 10], stddev=0.14),name="w4")
b4= tf.Variable(tf.constant(0.1,shape=[10]),name="b4")
y_conv = tf.matmul(fp3, w4) + b4 #定义交叉熵
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_,logits=y_conv))
#定义训练
train_step = tf.train.AdamOptimizer(1e-4).minimize (cross_entropy)
# 使用Dropout，keep_prob是一个占位符，训练时为0.5，测试时为1
keep_prob = tf.placeholder(tf.float32)
# 定义测试的准确率
correct_prediction = tf.equal(tf.argmax(y_conv, 1),tf.argmax(y_, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32)) # 创建Session和变量初始化
sess = tf.InteractiveSession()
sess.run (tf.global_variables_initializer()) #准确标志
maxaccuracy=0 #模型存储
saver = tf.train.Saver(max_to_keep=1)
# 训练1000步
for i in range (400):
batch = mnist.train.next_batch(50)
print(batch[0].shape)
if i % 100 == 0:
train_accuracy = accuracy.eval (feed_dict={x: batch[0], y_: batch[1], keep_prob: 1.0})
print ("step %d, training accuracy %g" % (i, train_accuracy))
train_step.run (feed_dict={x: batch[0], y_: batch[1], keep_prob: 0.5})
if maxaccuracy<train_accuracy:
maxaccuracy=train_accuracy
saver.save (sess, 'cnn/model.ckpt')
print("Cnn model training end..")
# print("test accuracy %g" % accuracy.eval (feed_dict={x: mnist.test.images, y_: mnist.test.labels, keep_prob: 1.0})) def predict():
meta_path = 'ckpt/mnist.ckpt.meta'
model_path = 'ckpt/mnist.ckpt'
sess = tf.InteractiveSession ()
saver = tf.train.import_meta_graph (meta_path)
saver.restore (sess, model_path)
graph = tf.get_default_graph ()
W = graph.get_tensor_by_name ("w:0")
b = graph.get_tensor_by_name ("b:0")
x = tf.placeholder (tf.float32, [None, 784])
y = tf.nn.softmax (tf.matmul (x, W) + b)
keep_prob = tf.placeholder (tf.float32)
batch_xs, batch_ys=mnist.train.next_batch (100)
one_img = batch_xs[0].reshape ((1, 784))
one_num = batch_ys[0].reshape ((1, 10))
temp = sess.run (y, feed_dict={x: one_img, keep_prob: 1.0})
b = sess.run (tf.argmax (temp, 1))
a = sess.run (tf.arg_max (one_num, 1))
print(temp)
print(one_num)
if b == a:
print ("success! the num is :", (b[0]))
showImgTest(one_img)
else:
print ("mistakes predict.") def trainNet():
x = tf.placeholder (tf.float32, [None, 784])
W = tf.Variable (tf.zeros ([784, 10]),name="w")
b = tf.Variable (tf.zeros ([10]),name="b")
y = tf.nn.softmax (tf.matmul (x, W) + b)
y_ = tf.placeholder (tf.float32, [None, 10])
keep_prob = tf.placeholder (tf.float32)
# 定义测试的准确率
correct_prediction = tf.equal (tf.argmax (y, 1), tf.argmax (y_, 1))
accuracy = tf.reduce_mean (tf.cast (correct_prediction, tf.float32))
#模型保存
saver = tf.train.Saver (max_to_keep=1)
#保存最优模型
max_acc=0
train_accuracy=0
#交叉熵
cross_entropy = tf.reduce_mean (-tf.reduce_sum (y_ * tf.log (y)))
# cross_error=cross_entropy_error_batch(y,y_)
train_step = tf.train.GradientDescentOptimizer (0.01).minimize (cross_entropy)
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
for i in range (1000):
batch_xs, batch_ys = mnist.train.next_batch (100)
sess.run (train_step, feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 1.0})
if i % 100 == 0:
train_accuracy = accuracy.eval (feed_dict={x: batch_xs, y_: batch_ys, keep_prob: 1.0})
print ("step %d, training accuracy %g" % (i, train_accuracy))
if train_accuracy > max_acc:
max_acc = train_accuracy
saver.save (sess, 'ckpt/mnist.ckpt') if __name__ == '__main__':
mnist = input_data.read_data_sets ("MNIST_data/", one_hot=True)
choice="0"
while choice == "0":
print ("------------------------tensorflow--------------------------")
print ("\t\t\t1\ttrain model..")
print("\t\t\t2\tpredict model")
print("\t\t\t3\tshow the first image")
print("\t\t\t4\tCNN model")
print("\t\t\t5\tpredict with cnn model")
print("\t\t\t6\tface recognized train with cnn")
print("\t\t\t7\tpredict tianjingle")
print("-------------------------------------------------------------")
print ("\t\t\t0\texit")
choice = input ("please input your choice！")
if choice == "1":
print("start train...")
trainNet()
if choice=="2":
predict()
if choice=="3":
showImg()
if choice=="4":
CNNmodel()
if choice=="5":
predictwithCnn()
if choice=="6":
trainface()
if choice=="7":
predicttjl() # train_x,train_y=mnist.train.next_batch(60000)
# print(train_x.shape,train_y.shape)
# img=train_x[1]
# k=img.reshape(28,28)
# #showImg(k)
# y=[0.1,0.05,0.6,0.0,0.05,0.1,0.0,0.1,0.0,0.0]
# t=[0,0,1,0,0,0,0,0,0,0]
# print(mean_squared_error(np.array(y),np.array(t))) #取图像的个数
