经典卷积神经网络---VGG16详解

水深无声 2023-10-04 11:28 194阅读 0赞

## 一.VGG概述 ##

VGGNet是牛津大学视觉几何组(Visual Geometry Group)提出的模型，该模型在2014ImageNet图像分类与定位挑战赛 ILSVRC-2014中取得在分类任务第二，定位任务第一的优异成绩。VGGNet突出的贡献是证明了很小的卷积，通过增加网络深度可以有效提高性能。VGG很好的继承了Alexnet的衣钵同时拥有着鲜明的特点。即网络层次较深。

**VGGNet结构**

VGGNet模型有A-E五种结构网络，深度分别为11,11,13,16,19。其中较为典型的网络结构主要有vgg16和vgg19，本篇文章主要讲VGG16,并分享VGG16的Keras实现。其网络结构如下图中D列（红色方框）：

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz_size_16_color_FFFFFF_t_70][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz_size_16_color_FFFFFF_t_70 1][]

**VGG16网络结构**

**vggnet对输入图像的默认大小是224\*224\*3 （从表中input可以看出）。vgg16网络结构含有参数的网络层一共有16层，即13个卷积层，5个池化层，3个全连接层，不包括激活层。**

**vgg16网络结构可以划分为6个模块层次加1个输入模块，分别如下**

<table style="width:500px;"> 
 <tbody> 
  <tr> 
   <td>&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;<strong> &nbsp; &nbsp;模块</strong></td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; <strong>&nbsp; &nbsp;各模块的涉及的层次</strong></td> 
  </tr> 
  <tr> 
   <td>&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 输入模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;&nbsp;<strong>224*224*3</strong></td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 第一个模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-64</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-64</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; maxpool</td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 第二个模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-128</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-128</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; maxpool</td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;第三个模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-256</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-256</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-256</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; maxpool</td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 第四个模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; maxpool</td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp;第五个模块</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; conv3-512</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; maxpool</td> 
  </tr> 
  <tr> 
   <td colspan="1">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; 第六个模块（全连接层和输出层）</td> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; FC-4096 （实际上前面需要加一个Flatten层）</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; FC-4096</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; FC-1000 (负责分类)</td> 
  </tr> 
  <tr> 
   <td style="width:275px;">&nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; &nbsp; softmax(输出层函数)</td> 
  </tr> 
 </tbody> 
</table>

## 二.vgg16实现MNIST分类 ##

（基于keras框架）

**代码实现：**

#从keras.model中导入model模块，为函数api搭建网络做准备
    from keras.models import Model
    from keras.layers import Flatten,Dense,Dropout,MaxPooling2D,Conv2D,BatchNormalization,Input,ZeroPadding2D,Concatenate
    from keras.layers.convolutional import AveragePooling2D
    from keras import regularizers  #正则化
    from keras.optimizers import RMSprop  #优化选择器
    from keras.layers import AveragePooling2D
    from keras.datasets import mnist
    from keras.utils import np_utils
    import matplotlib.pyplot as plt
    import numpy as np
    
    #数据处理
    (X_train,Y_train),(X_test,Y_test)=mnist.load_data()
    X_test1=X_test
    Y_test1=Y_test
    X_train=X_train.reshape(-1,28,28,1).astype("float32")/255.0
    X_test=X_test.reshape(-1,28,28,1).astype("float32")/255.0
    Y_train=np_utils.to_categorical(Y_train,10)
    Y_test=np_utils.to_categorical(Y_test,10)
    print(X_train.shape)
    print(Y_train.shape)
    print(X_train.shape)
    
    def vgg16():
        x_input = Input((28, 28, 1))  # 输入数据形状28*28*1
        # Block 1
        x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv1')(x_input)
        x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x)
        x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
    
        # Block 2
        x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x)
        x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x)
        x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)
    
        # Block 3
        x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x)
        x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv2')(x)
        x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv3')(x)
        x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool')(x)
    
        # Block 4
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv1')(x)
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv2')(x)
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv3')(x)
        x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool')(x)
    
        # Block 5
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv1')(x)
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv2')(x)
        x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv3')(x)
    
        #BLOCK 6
        x=Flatten()(x)
        x=Dense(256,activation="relu")(x)
        x=Dropout(0.5)(x)
        x = Dense(256, activation="relu")(x)
        x = Dropout(0.5)(x)
        #搭建最后一层，即输出层
        x = Dense(10, activation="softmax")(x)
        # 调用MDOEL函数，定义该网络模型的输入层为X_input,输出层为x.即全连接层
        model = Model(inputs=x_input, outputs=x)
        # 查看网络模型的摘要
        model.summary()
        return model
    model=vgg16()
    optimizer=RMSprop(lr=1e-4)
    model.compile(loss="binary_crossentropy",optimizer=optimizer,metrics=["accuracy"])
    #训练加评估模型
    n_epoch=4
    batch_size=128
    def run_model(): #训练模型
        training=model.fit(
        X_train,
        Y_train,
        batch_size=batch_size,
        epochs=n_epoch,
        validation_split=0.25,
        verbose=1
        )
        test=model.evaluate(X_train,Y_train,verbose=1)
        return training,test
    training,test=run_model()
    print("误差：",test[0])
    print("准确率：",test[1])
    
    def show_train(training_history,train, validation):
        plt.plot(training.history[train],linestyle="-",color="b")
        plt.plot(training.history[validation] ,linestyle="--",color="r")
        plt.title("training history")
        plt.xlabel("epoch")
        plt.ylabel("accuracy")
        plt.legend(["training","validation"],loc="lower right")
        plt.show()
    show_train(training,"accuracy","val_accuracy")
    
    def show_train1(training_history,train, validation):
        plt.plot(training.history[train],linestyle="-",color="b")
        plt.plot(training.history[validation] ,linestyle="--",color="r")
        plt.title("training history")
        plt.xlabel("epoch")
        plt.ylabel("loss")
        plt.legend(["training","validation"],loc="upper right")
        plt.show()
    show_train1(training,"loss","val_loss")
    
    prediction=model.predict(X_test)
    def image_show(image):
        fig=plt.gcf()  #获取当前图像
        fig.set_size_inches(2,2)  #改变图像大小
        plt.imshow(image,cmap="binary")  #显示图像
        plt.show()
    def result(i):
        image_show(X_test1[i])
        print("真实值：",Y_test1[i])
        print("预测值：",np.argmax(prediction[i]))
    result(0)
    result(1)

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20210421153431106.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20210421154043200.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2hnbnV4Y18xOTkz,size_16,color_FFFFFF,t_70