Tensorflow 迁移学习[实现]

当我们要使用神经网络解决某一问题时，有时我们并不需要从0开始训练一个网络，因为训练一个较为复杂的网络，不仅会消耗很多时间也需要较大的数据集。

我们可以使用别人在解决与我们类似的问题时，已经训练好的网络和相应的参数，将网络结构进行“微调”，使得它适合我们的问题，最后在我们的数据集上进行再次的训练。

例如，在解决图像分类问题时
我们可以使用已经在大型的数据集上训练好的InceptionV3网络。

由于卷积层的主要作用是对图像特征的提取，并且在前几层卷积层，提取的方法（权重）基本相同。都是通过寻找图片中的 边，线，轮廓，然后组合。所以在使用InceptionV3时，我们可以锁住前面的卷积层的权重（训练时不进行更新），仅仅对后面的全连接层的参数进行训练。不仅可以保证准确率还能提高训练所需的时间。

---

实现：
（1）预定义InceptionV3网络，加载参数，锁住卷积层，或者卷积层的输出

import os

from tensorflow.keras import layers
from tensorflow.keras import Model
#下载网络参数，放置在 /tmp/
!wget --no-check-certificate \
    https://storage.googleapis.com/mledu-datasets/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5 \
    -O /tmp/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5
  
from tensorflow.keras.applications.inception_v3 import InceptionV3

local_weights_file = '/tmp/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5'
#预定义InceptionV3网络
pre_trained_model = InceptionV3(input_shape = (150, 150, 3), #输入150x150的彩色图片
                                include_top = False, #去除网络卷积层前面的全连接层
                                weights = None) #无参数，稍后加载

pre_trained_model.load_weights(local_weights_file)  #加载预训练参数

for layer in pre_trained_model.layers:
  layer.trainable = False #锁住所有的InceptionV3网络层权重
  
# pre_trained_model.summary()

last_layer = pre_trained_model.get_layer('mixed7') #获得InceptionV3网络mixed7卷积层
print('last layer output shape: ', last_layer.output_shape)
last_output = last_layer.output #获得卷积层的输出

（2）在卷积层后，加入我们自己所需的网络结构

from tensorflow.keras.optimizers import RMSprop

# 展开卷积层的输出
x = layers.Flatten()(last_output)
# 加入一个有1024个神经元的隐藏层，使用relu激活函数
x = layers.Dense(1024, activation='relu')(x)
# 防止过拟合，使用Dropout随机丢弃20%的节点
x = layers.Dropout(0.2)(x)                  
# 输出层
x = layers.Dense  (1, activation='sigmoid')(x)           
#使用预训练模型的输入和我们自己定义的输出，生成最终的网络模型
model = Model( pre_trained_model.input, x) 
#设置优化器，损失函数等
model.compile(optimizer = RMSprop(lr=0.0001), 
              loss = 'binary_crossentropy', 
              metrics = ['acc'])

（3）网络结构定义好后，接下来就只需进行，下载数据集进行网络的训练了。

!wget --no-check-certificate \
        https://storage.googleapis.com/mledu-datasets/cats_and_dogs_filtered.zip \
       -O /tmp/cats_and_dogs_filtered.zip

from tensorflow.keras.preprocessing.image import ImageDataGenerator

import os
import zipfile

local_zip = '//tmp/cats_and_dogs_filtered.zip'

zip_ref = zipfile.ZipFile(local_zip, 'r')

zip_ref.extractall('/tmp')
zip_ref.close()

# Define our example directories and files
base_dir = '/tmp/cats_and_dogs_filtered'

train_dir = os.path.join( base_dir, 'train')
validation_dir = os.path.join( base_dir, 'validation')


train_cats_dir = os.path.join(train_dir, 'cats') # Directory with our training cat pictures
train_dogs_dir = os.path.join(train_dir, 'dogs') # Directory with our training dog pictures
validation_cats_dir = os.path.join(validation_dir, 'cats') # Directory with our validation cat pictures
validation_dogs_dir = os.path.join(validation_dir, 'dogs')# Directory with our validation dog pictures

train_cat_fnames = os.listdir(train_cats_dir)
train_dog_fnames = os.listdir(train_dogs_dir)

# Add our data-augmentation parameters to ImageDataGenerator
train_datagen = ImageDataGenerator(rescale = 1./255.,
                                   rotation_range = 40,
                                   width_shift_range = 0.2,
                                   height_shift_range = 0.2,
                                   shear_range = 0.2,
                                   zoom_range = 0.2,
                                   horizontal_flip = True)

# Note that the validation data should not be augmented!
test_datagen = ImageDataGenerator( rescale = 1.0/255. )

# Flow training images in batches of 20 using train_datagen generator
train_generator = train_datagen.flow_from_directory(train_dir,
                                                    batch_size = 20,
                                                    class_mode = 'binary', 
                                                    target_size = (150, 150))     

# Flow validation images in batches of 20 using test_datagen generator
validation_generator =  test_datagen.flow_from_directory( validation_dir,
                                                          batch_size  = 20,
                                                          class_mode  = 'binary', 
                                                          target_size = (150, 150))

history = model.fit_generator(
            train_generator,
            validation_data = validation_generator,
            steps_per_epoch = 100,
            epochs = 20,
            validation_steps = 50,
            verbose = 2)

import matplotlib.pyplot as plt
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']

epochs = range(len(acc))

plt.plot(epochs, acc, 'r', label='Training accuracy')
plt.plot(epochs, val_acc, 'b', label='Validation accuracy')
plt.title('Training and validation accuracy')
plt.legend(loc=0)
plt.figure()


plt.show()