AIR

利用深度神经网络进行基本图像分类

此代码在colab已经经过测试并且可以查看TensorBoard：地址

环境：Tensorflow-2.1.0、Python3.6、google colab

In[1]:

1
2
3
import ssl
ssl._create_default_https_context = ssl._create_unverified_context
# 不知道为什么在本地的jupyter notebook上无法导入fashion_mnist数据集(本地需要，google colab不需要)

In[2]:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
import matplotlib as mpl
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import sklearn 
import pandas as pd
import os
import sys
import time
import tensorflow as tf

from tensorflow import keras as keras
from sklearn.preprocessing import StandardScaler

for module in mpl,np,pd,sklearn,tf,keras:
    print(module.__name__,module.__version__)
# 打印版本库

Out[1]:

matplotlib 3.1.2
numpy 1.18.1
pandas 0.25.3
sklearn 0.22.1
tensorflow 2.1.0
tensorflow_core.keras 2.2.4-tf

In[3]:

1
2
3
fashion_mnist=keras.datasets.fashion_mnist
(x_train_all, y_train_all),(x_test, y_test)=fashion_mnist.load_data()
# 导入fashion_mnist数据集

In[4]

1
2
3
4
5
6
7
x_valid,x_train=x_train_all[:5000],x_train_all[5000:]
y_valid,y_train=y_train_all[:5000],y_train_all[5000:]
# fashion_mnist有6w个数据，前5k个为验证集，其中x为图片，y为序列号
print(x_valid.shape,y_valid.shape)
print(x_train.shape,y_valid.shape)
print(x_test.shape,y_test.shape)
# 打印数据集的格式

Out[2]

(5000, 28, 28) (5000,)
(55000, 28, 28) (5000,)
(10000, 28, 28) (10000,)

In[5]:

1
2
print(np.max(x_train),np.min(x_train))
# 打印数据集中的最大值和最小值

Out[3]:

255 0

In[6]:

1
2
3
4
5
6
7
8
9
# x= (x-u)/std
scaler=StandardScaler()
# x_train:[None, 28, 28] -> [None, 784]
x_train_scaled=scaler.fit_transform(x_train.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
#会自动记录方差u和均值std，并进行归一化
x_valid_scaled=scaler.transform(x_valid.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
# 会根据自动记录的方差u和均值std进行归一化
x_test_scaled=scaler.transform(x_test.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
# 会根据自动记录的方差u和均值std进行归一化

Out[4]:

2.0231433 -0.8105136

In[7]:

1
2
3
4
5
6
def show_single_image(img_arr):
    plt.imshow(img_arr,cmap='binary')
    plt.show()
# 打印单张图片
    
show_single_image(x_train[0])

Out[5]:

In[8]:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
def show_images(n_rows,n_cols,x_data,y_data,class_names):
    assert len(x_data)==len(y_data)
    assert n_cols*n_rows<len(x_data)
    plt.figure(figsize=(n_cols*1.4,n_rows*1.6))
    for row in range(n_rows):
        for col in range(n_cols):
            index=n_cols*row+col
            plt.subplot(n_rows,n_cols,index+1)
            plt.imshow(x_data[index],cmap='binary',
                       interpolation='nearest')
            plt.axis('off')
            plt.title(class_names[y_data[index]])
    plt.show()
# 打印n_rows行n_cols列的图片
    
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# 所对应的名称

show_images(3,5,x_train,y_train,class_names)

Out[6]:

In[9]:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
# 构建一个三层的神经网络
# model=keras.models.Sequential()
# model.add(keras.layers.Flatten(input_shape=[28,28])) # 展开成28*28的向量
# model.add(keras.layers.Dense(300,activation='relu')) # 全链接层，层数300，激活函数relu
# model.add(keras.layers.Dense(100,activation='relu')) 
# model.add(keras.layers.Dense(10,activation='softmax')) # 有10个输出

# model=keras.models.Sequential([
#     keras.layers.Flatten(input_shape=[28,28]),
#     keras.layers.Dense(300,activation='relu'),
#     keras.layers.Dense(100,activation='relu'),
#     keras.layers.Dense(10,activation='softmax')
# ])

model=keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28,28]))
for _ in range(20):
    model.add(keras.layers.Dense(100,activation='selu'))
    
    # model.add(keras.layers.Dense(100,activation='relu'))
    # model.add(keras.layers.BatchNormalization())
    
    # 激活函数在前的后的批归一化
    # model.add(keras.layers.Dense(100))
    # model.add(keras.layers.BatchNormalization()) 
    # model.add(keras.layers.Activation('relu'))
    
# model.add(keras.layers.Dropout(rate=0.5))
model.add(keras.layers.AlphaDropout(rate=0.5))
# AlphaDropout与Dropout的区别：1、均值方差不变 2、归一化性质不变
model.add(keras.layers.Dense(10,activation='softmax'))

# relu: y=max(0,x)
# softmax: 将向量变成概率分布 x=[x1,x2,x3]
#          y=[e^x1/sum,e^x2/sum,e^x3/sum],sum=e^x1+e^x2+e^x3

# reason for sparse: y->index y->one_hot->[]

model.compile(loss='sparse_categorical_crossentropy',
              optimizer='sgd',
              metrics=['accuracy'])

tensorflow.keras.model.compile

In[10]:

1
model.layers

Out[7]:

[<tensorflow.python.keras.layers.core.Flatten at 0x136ef6a90>,
 <tensorflow.python.keras.layers.core.Dense at 0x136ef6ad0>,
 <tensorflow.python.keras.layers.core.Dense at 0x136eb5f90>,
 <tensorflow.python.keras.layers.core.Dense at 0x13689dcd0>,
 <tensorflow.python.keras.layers.core.Dense at 0x136ed6990>,
 <tensorflow.python.keras.layers.core.Dense at 0x136894090>,
 <tensorflow.python.keras.layers.core.Dense at 0x13689b450>,
 <tensorflow.python.keras.layers.core.Dense at 0x13689b110>,
 <tensorflow.python.keras.layers.core.Dense at 0x111477050>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a68c7d0>,
 <tensorflow.python.keras.layers.core.Dense at 0x1114bd810>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a7035d0>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a6a9050>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a6a9bd0>,
 <tensorflow.python.keras.layers.core.Dense at 0x107df4650>,
 <tensorflow.python.keras.layers.core.Dense at 0x107e2b410>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a724910>,
 <tensorflow.python.keras.layers.core.Dense at 0x13a6dd210>,
 <tensorflow.python.keras.layers.core.Dense at 0x13e4b1b90>,
 <tensorflow.python.keras.layers.core.Dense at 0x137080b10>,
 <tensorflow.python.keras.layers.core.Dense at 0x13e4a1a10>,
 <tensorflow.python.keras.layers.noise.AlphaDropout at 0x136ed69d0>,
 <tensorflow.python.keras.layers.core.Dense at 0x1370af710>]

In[11]:

1
model.summary() #打印模型的结构

Out[8]:

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten (Flatten)            (None, 784)               0         
_________________________________________________________________
dense (Dense)                (None, 100)               78500     
_________________________________________________________________
dense_1 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_2 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_3 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_4 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_5 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_6 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_7 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_8 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_9 (Dense)              (None, 100)               10100     
_________________________________________________________________
dense_10 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_11 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_12 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_13 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_14 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_15 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_16 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_17 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_18 (Dense)             (None, 100)               10100     
_________________________________________________________________
dense_19 (Dense)             (None, 100)               10100     
_________________________________________________________________
alpha_dropout (AlphaDropout) (None, 100)               0         
_________________________________________________________________
dense_20 (Dense)             (None, 10)                1010      
=================================================================
Total params: 271,410
Trainable params: 271,410
Non-trainable params: 0
_________________________________________________________________

In[12]:

1
# 第一层 [None, 784] * w + b -> [None, 300] w.shape=[784,300], b=[300]

In[13]:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
# TensorBoard,EarlyStopping,ModelCheckPoint
logdir='./dnn-selu-dropout-callbacks' # 设置一个文件夹路径
if not os.path.exists(logdir): # 若文件夹不存在，则创建文件夹
    os.mkdir(logdir)
output_model_file=os.path.join(logdir,
                               'fashion_mnist_model.h5') #放入文件夹内，文件名fashion_mnist_model.h5
callbacks=[
    keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,
                                    save_best_only=True), # 保存最好的模型，如果不是这个会默认保存最近的一个模型
    keras.callbacks.EarlyStopping(patience=5,min_delta=1e-3)
]
history=model.fit(x_train_scaled,y_train,epochs=10,
          validation_data=(x_valid_scaled,y_valid),
                  callbacks=callbacks)

tensorflow.keras.model.fit

Out[9]:

Train on 55000 samples, validate on 5000 samples
Epoch 1/10
55000/55000 [==============================] - 11s 195us/sample - loss: 0.6998 - accuracy: 0.7617 - val_loss: 0.5656 - val_accuracy: 0.8536
Epoch 2/10
55000/55000 [==============================] - 8s 144us/sample - loss: 0.4626 - accuracy: 0.8428 - val_loss: 0.6195 - val_accuracy: 0.8558
Epoch 3/10
55000/55000 [==============================] - 9s 162us/sample - loss: 0.4065 - accuracy: 0.8581 - val_loss: 0.5300 - val_accuracy: 0.8668
Epoch 4/10
55000/55000 [==============================] - 8s 142us/sample - loss: 0.3755 - accuracy: 0.8717 - val_loss: 0.6154 - val_accuracy: 0.8664
Epoch 5/10
55000/55000 [==============================] - 8s 152us/sample - loss: 0.3521 - accuracy: 0.8774 - val_loss: 0.5287 - val_accuracy: 0.8766
Epoch 6/10
55000/55000 [==============================] - 8s 153us/sample - loss: 0.3343 - accuracy: 0.8815 - val_loss: 0.4956 - val_accuracy: 0.8856
Epoch 7/10
55000/55000 [==============================] - 8s 143us/sample - loss: 0.3193 - accuracy: 0.8857 - val_loss: 0.5349 - val_accuracy: 0.8716
Epoch 8/10
55000/55000 [==============================] - 8s 144us/sample - loss: 0.3085 - accuracy: 0.8916 - val_loss: 0.5294 - val_accuracy: 0.8842
Epoch 9/10
55000/55000 [==============================] - 8s 145us/sample - loss: 0.2955 - accuracy: 0.8946 - val_loss: 0.4668 - val_accuracy: 0.8822
Epoch 10/10
55000/55000 [==============================] - 8s 152us/sample - loss: 0.2839 - accuracy: 0.8975 - val_loss: 0.4907 - val_accuracy: 0.8860

In[14]:

1
type(history)

Out[10]:

tensorflow.python.keras.callbacks.History

In[15]:

1
2
history.history
# 打印每次学习的结果

Out[11]:

{'loss': [0.6997782779043371,
  0.4625766654621471,
  0.4064591264659708,
  0.37553091496554286,
  0.35207060631621967,
  0.3343261710730466,
  0.3193407494176518,
  0.30854321794509887,
  0.29552189791636035,
  0.28390346760099583],
 'accuracy': [0.76165456,
  0.84276366,
  0.85809094,
  0.8717273,
  0.87743634,
  0.8814909,
  0.8856909,
  0.89156365,
  0.89461815,
  0.8974909],
 'val_loss': [0.5656432386219501,
  0.6194551657661795,
  0.5299783393919468,
  0.6153697894215584,
  0.5287225459891837,
  0.49561948866695166,
  0.5348633618405322,
  0.5294362505242228,
  0.4668049536377192,
  0.4907300000913441],
 'val_accuracy': [0.8536,
  0.8558,
  0.8668,
  0.8664,
  0.8766,
  0.8856,
  0.8716,
  0.8842,
  0.8822,
  0.886]}

In[16]

1
2
3
4
5
6
7
8
def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize=(8,5))
    plt.grid(True)
    plt.gca().set_ylim(0,1)
    plt.show()
# 打印学习曲线

plot_learning_curves(history)

Out[12]:

In[17]:

1
model.evaluate(x_test_scaled,y_test) # 在测试集上验证

tensorflow.keras.model.evaluate

Out[13]:

10000/10000 [==============================] - 1s 95us/sample - loss: 0.5799 - accuracy: 0.8741

Out[14]:

[0.5799049034297467, 0.8741]

汇总：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
#!/usr/bin/env python
# coding: utf-8

# In[1]:


import ssl
ssl._create_default_https_context = ssl._create_unverified_context
# 不知道为什么在本地的jupyter notebook上无法导入fashion_mnist数据集(本地需要，google colab不需要)


# In[23]:


import matplotlib as mpl
import matplotlib.pyplot as plt
get_ipython().run_line_magic('matplotlib', 'inline')
import numpy as np
import sklearn 
import pandas as pd
import os
import sys
import time
import tensorflow as tf

from tensorflow import keras as keras
from sklearn.preprocessing import StandardScaler

for module in mpl,np,pd,sklearn,tf,keras:
    print(module.__name__,module.__version__)
# 打印版本库


# In[3]:


fashion_mnist=keras.datasets.fashion_mnist
(x_train_all, y_train_all),(x_test, y_test)=fashion_mnist.load_data()
# 导入fashion_mnist数据集


# In[4]:


x_valid,x_train=x_train_all[:5000],x_train_all[5000:]
y_valid,y_train=y_train_all[:5000],y_train_all[5000:]
# fashion_mnist有6w个数据，前5k个为验证集，其中x为图片，y为序列号
print(x_valid.shape,y_valid.shape)
print(x_train.shape,y_valid.shape)
print(x_test.shape,y_test.shape)
# 打印数据集的格式


# In[5]:


print(np.max(x_train),np.min(x_train))
# 打印数据集中的最大值和最小值


# In[6]:


# x= (x-u)/std
scaler=StandardScaler()
# x_train:[None, 28, 28] -> [None, 784]
x_train_scaled=scaler.fit_transform(x_train.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
#会自动记录方差u和均值std，并进行归一化
x_valid_scaled=scaler.transform(x_valid.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
# 会根据自动记录的方差u和均值std进行归一化
x_test_scaled=scaler.transform(x_test.astype(np.float32).reshape(-1,1)).reshape(-1,28,28) 
# 会根据自动记录的方差u和均值std进行归一化


# In[7]:


def show_single_image(img_arr):
    plt.imshow(img_arr,cmap='binary')
    plt.show()
# 打印单张图片
    
show_single_image(x_train[0])


# In[8]:


def show_images(n_rows,n_cols,x_data,y_data,class_names):
    assert len(x_data)==len(y_data)
    assert n_cols*n_rows<len(x_data)
    plt.figure(figsize=(n_cols*1.4,n_rows*1.6))
    for row in range(n_rows):
        for col in range(n_cols):
            index=n_cols*row+col
            plt.subplot(n_rows,n_cols,index+1)
            plt.imshow(x_data[index],cmap='binary',
                       interpolation='nearest')
            plt.axis('off')
            plt.title(class_names[y_data[index]])
    plt.show()
# 打印n_rows行n_cols列的图片
    
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
               'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
# 所对应的名称

show_images(3,5,x_train,y_train,class_names)


# In[9]:


# 构建一个三层的神经网络
# model=keras.models.Sequential()
# model.add(keras.layers.Flatten(input_shape=[28,28])) # 展开成28*28的向量
# model.add(keras.layers.Dense(300,activation='relu')) # 全链接层，层数300，激活函数relu
# model.add(keras.layers.Dense(100,activation='relu')) 
# model.add(keras.layers.Dense(10,activation='softmax')) # 有10个输出

# model=keras.models.Sequential([
#     keras.layers.Flatten(input_shape=[28,28]),
#     keras.layers.Dense(300,activation='relu'),
#     keras.layers.Dense(100,activation='relu'),
#     keras.layers.Dense(10,activation='softmax')
# ])

model=keras.models.Sequential()
model.add(keras.layers.Flatten(input_shape=[28,28]))
for _ in range(20):
    model.add(keras.layers.Dense(100,activation='selu'))
    
    # model.add(keras.layers.Dense(100,activation='relu'))
    # model.add(keras.layers.BatchNormalization())
    
    # 激活函数在前的后的批归一化
    # model.add(keras.layers.Dense(100))
    # model.add(keras.layers.BatchNormalization()) 
    # model.add(keras.layers.Activation('relu'))
    
# model.add(keras.layers.Dropout(rate=0.5))
model.add(keras.layers.AlphaDropout(rate=0.5))
# AlphaDropout与Dropout的区别：1、均值方差不变 2、归一化性质不变
model.add(keras.layers.Dense(10,activation='softmax'))

# relu: y=max(0,x)
# softmax: 将向量变成概率分布 x=[x1,x2,x3]
#          y=[e^x1/sum,e^x2/sum,e^x3/sum],sum=e^x1+e^x2+e^x3

# reason for sparse: y->index y->one_hot->[]

model.compile(loss='sparse_categorical_crossentropy',
              optimizer='sgd',
              metrics=['accuracy'])
    


# In[10]:


model.layers


# In[11]:


model.summary() #打印模型的结构


# In[12]:


# 第一层 [None, 784] * w + b -> [None, 300] w.shape=[784,300], b=[300]


# In[15]:


# TensorBoard,EarlyStopping,ModelCheckPoint
logdir='./dnn-selu-dropout-callbacks' # 设置一个文件夹路径
if not os.path.exists(logdir): # 若文件夹不存在，则创建文件夹
    os.mkdir(logdir)
output_model_file=os.path.join(logdir,
                               'fashion_mnist_model.h5') #放入文件夹内，文件名fashion_mnist_model.h5
callbacks=[
    keras.callbacks.TensorBoard(logdir),
    keras.callbacks.ModelCheckpoint(output_model_file,
                                    save_best_only=True), # 保存最好的模型，如果不是这个会默认保存最近的一个模型
    keras.callbacks.EarlyStopping(patience=5,min_delta=1e-3)
]
history=model.fit(x_train_scaled,y_train,epochs=10,
          validation_data=(x_valid_scaled,y_valid),
                  callbacks=callbacks)


# In[16]:


type(history)


# In[17]:


history.history
# 打印每次学习的结果


# In[18]:


def plot_learning_curves(history):
    pd.DataFrame(history.history).plot(figsize=(8,5))
    plt.grid(True)
    plt.gca().set_ylim(0,1)
    plt.show()
# 打印学习曲线

plot_learning_curves(history)
    


# In[19]:


model.evaluate(x_test_scaled,y_test) # 在测试集上验证
本文由USC-AIR创作和发表,采用BY-NC-SA国际许可协议进行许可
转载请注明作者及出处,本文作者为USC-AIR,本文标题为利用深度神经网络进行基本图像分类
本文链接为http://uscair.club/2020/02/22/利用深度神经网络进行基本图像分类/.

 Previous

STM32F407

 Comments


Blog content follows the Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) License

Use Material X as theme , total visits times .
载入天数...载入时分秒...