Keras实现vgg16网络和迁移学习

VGGnet

VGGnet是由牛津大学和DeepMind研发的深度学习网络(2014年)。它是由Alexnet发展而来的，其中最为经典的是vgg16和vgg19 ,至今任被广泛应用于图像特征提取

下面是vgg11-vgg19的结构，vgg后的数字代表网络层数

Keras定义vgg16

vgg16结构：

上图的输入是224*224，可以根据实际做调整

keras使我们很容易构建深度学习模型

根据上图定义的vgg16模型：

from keras.models import Sequential
from keras.layers import Conv2D,Dense,Flatten,Dropout,MaxPool2D,BatchNormalization

def vgg16_model():
    model=Sequential()

    model.add(Conv2D(filters=64,kernel_size=(3,3),strides=(1,1),input_shape=(224,224,3),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(filters=64,kernel_size=(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(MaxPool2D()) # pool_size=(2,2) strides=(2,2)

    model.add(Conv2D(128,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(128,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(MaxPool2D())

    model.add(Conv2D(256,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(256,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(256,(1,1),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(MaxPool2D())

    model.add(Conv2D(512,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(512,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(512,(1,1),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(MaxPool2D())

    model.add(Conv2D(512,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(512,(3,3),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(Conv2D(512,(1,1),strides=(1,1),padding='same',activation='relu',kernel_initializer='glorot_uniform'))
    model.add(MaxPool2D())

    model.add(Flatten())
    model.add(Dense(4096,activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(4096,activation='relu'))
    model.add(Dropout(0.5))
    model.add(Dense(1000,activation='relu'))

    model.add(Dropout(0.5))
    model.add(Dense(1000,activation='softmax'))

    return model

model=vgg16_model()
model.summary()

_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_146 (Conv2D)         (None, 224, 224, 64)      1792      
                                                                 
 conv2d_147 (Conv2D)         (None, 224, 224, 64)      36928     
                                                                 
 max_pooling2d_56 (MaxPoolin  (None, 112, 112, 64)     0         
 g2D)                                                            
                                                                 
 conv2d_148 (Conv2D)         (None, 112, 112, 128)     73856     
                                                                 
 conv2d_149 (Conv2D)         (None, 112, 112, 128)     147584    
                                                                 
 max_pooling2d_57 (MaxPoolin  (None, 56, 56, 128)      0         
 g2D)                                                            
                                                                 
 conv2d_150 (Conv2D)         (None, 56, 56, 256)       295168    
                                                                 
 conv2d_151 (Conv2D)         (None, 56, 56, 256)       590080    
                                                                 
 conv2d_152 (Conv2D)         (None, 56, 56, 256)       65792     
                                                                 
 max_pooling2d_58 (MaxPoolin  (None, 28, 28, 256)      0         
...
Total params: 134,639,952
Trainable params: 134,639,952
Non-trainable params: 0

Keras实现vgg16迁移学习

训练数据:

cifar10数据集（由 Hinton 的学生 Alex Krizhevsky 和 Ilya Sutskever 整理的一个用于识别普适物体的小型数据集。一共包含 10 个类别的 RGB 彩色图 片：飞机（ a叩lane ）、汽车（ automobile ）、鸟类（ bird ）、猫（ cat ）、鹿（ deer ）、狗（ dog ）、蛙类（ frog ）、马（ horse ）、船（ ship ）和卡车（ truck ）。图片的尺寸为 32×32 ，数据集中一共有 50000 张训练图片和 10000 张测试图片）

Keras.applications:

提供了许多带有预训练权值的深度学习模型

livelossplot:

一个绘制损失图像的库，可以在训练过程中实时绘制训练情况

from keras.datasets import cifar10
from keras.utils import to_categorical
from keras import applications
from keras.models import Sequential,Model
from keras.layers import Dense,Dropout,Flatten
from keras import optimizers

# 加载cifar10数据集
(x_train, y_train), (x_test, y_test) = cifar10.load_data() 
x_train = x_train.astype('float32') / 255  
x_test = x_test.astype('float32') / 255  
y_train = to_categorical(y_train, 10) #按照10个类别one-hot编码
y_test = to_categorical(y_test, 10)

# 加载keras训练后的vgg16模型
vgg_model= applications.VGG16(include_top=False,input_shape=(32,32,3))
for layer in vgg_model.layers[:15]: #冻结前15层
    layer.trainable=False

# 定义迁移学习层
top_model=Sequential()
top_model.add(Flatten(input_shape=vgg_model.output_shape[1:]))
top_model.add(Dense(32, activation='relu'))  
top_model.add(Dropout(0.5))  
top_model.add(Dense(10, activation='softmax')) 

# 结合预训练模型和迁移学习层为新模型
model = Model(
    inputs=vgg_model.input,
    outputs=top_model(vgg_model.output)) 

model.compile(
    loss='categorical_crossentropy',
    optimizer = optimizers.Adam(learning_rate=0.0001),
    metrics=['accuracy'])  

# 边训练边绘制损失
from livelossplot import PlotLossesKeras
plotlosses = PlotLossesKeras()
model.fit(x_train,y_train,epochs=5,verbose=1,validation_split=0.1,batch_size=32,callbacks=plotlosses)

训练中损失和分类准确率变化：

# 测试集损失与准确率
test_loss, test_acc = model.evaluate(x_test,y_test)
print('test_loss:{:.2} test_acc:{}%'.format(test_loss,test_acc*100))

# 保存模型
model.save('trained/cifar10.h5')

预测新图-识别猫咪

用学习出的模型预测猫咪

from keras.models import load_model
from keras.utils import image_utils
import matplotlib.pyplot as plt
import numpy as np

# 加载模型
model=load_model("trained/cifar10.h5")

# 图片预处理
path = 'data/my_cat.jpg'
img_height, img_width = 32, 32
x = image_utils.load_img(path=path, target_size=(img_height, img_width))
x = image_utils.img_to_array(x)
# print(x.shape) # (32, 32, 3)
x = x[None]      # 相当于增加一个维度
# print(x.shape) # (1, 32, 32, 3)

# 预测
y = model.predict(x)
labels=["airplane","automobile","bird","cat","deer","dog","frog","horse","ship","truck"]

result=[i.tolist().index(1) for i in y]
print("This is a {}".format(labels[result[0]]))

img = Image.open(path)
# img.show() # 会调用系统的显示窗口
plt.figure(dpi=120)
plt.imshow(img)

成功识别我的猫！