如何使用TensorFlow和预训练模型进行特征提取？

TensorFlow和预训练模型可以通过将之前创建的`base_model`的`trainable`属性设置为`False`来用于特征提取。

包含至少一层卷积层的神经网络被称为卷积神经网络。我们可以使用卷积神经网络来构建学习模型。

我们将了解如何借助来自预训练网络的迁移学习来对猫和狗的图像进行分类。图像分类迁移学习背后的直觉是，如果一个模型在一个大型且通用的数据集上进行训练，则此模型可以有效地用作视觉世界的通用模型。它将学习特征映射，这意味着用户不必从头开始在一个大型数据集上训练大型模型。

阅读更多： 如何预训练自定义模型？

我们使用Google Colaboratory运行以下代码。Google Colab或Colaboratory帮助在浏览器上运行Python代码，无需任何配置，并可免费访问GPU（图形处理单元）。Colaboratory构建在Jupyter Notebook之上。

示例

print("Feature extraction")
base_model.trainable = False
print("The base model architecture")
base_model.summary()

代码来源 −https://tensorflowcn.cn/tutorials/images/transfer_learning

输出

Feature extraction
The base model architecture
Model: "mobilenetv2_1.00_160"
__________________________________________________________________________________________________
Layer (type)                   Output Shape       Param #   Connected to
==================================================================================================
input_1 (InputLayer)         [(None, 160, 160, 3)             0
__________________________________________________________________________________________________
Conv1 (Conv2D)              (None, 80, 80, 32)      864       input_1[0][0]
__________________________________________________________________________________________________
bn_Conv1 (BatchNormalization) (None, 80, 80, 32)   128         Conv1[0][0]
__________________________________________________________________________________________________
Conv1_relu (ReLU)           (None, 80, 80, 32)       0       bn_Conv1[0][0]
__________________________________________________________________________________________________
expanded_conv_depthwise (Depthw (None, 80, 80, 32)   288         Conv1_relu[0][0]
__________________________________________________________________________________________________
expanded_conv_depthwise_BN (Bat (None, 80, 80, 32)   128       expanded_conv_depthwise[0][0]
__________________________________________________________________________________________________
expanded_conv_depthwise_relu (R (None, 80, 80, 32)    0       expanded_conv_depthwise_BN[0][0]
__________________________________________________________________________________________________
expanded_conv_project (Conv2D)  (None, 80, 80, 16)   512           expanded_conv_depthwise_relu[0][0
__________________________________________________________________________________________________
expanded_conv_project_BN (Batch (None, 80, 80, 16)      64       expanded_conv_project[0][0]
__________________________________________________________________________________________________
block_1_expand (Conv2D)    (None, 80, 80, 96)         1536         expanded_conv_project_BN[0][0]
__________________________________________________________________________________________________
block_1_expand_BN (BatchNormali  (None, 80, 80, 96)     384            block_1_expand[0][0]
__________________________________________________________________________________________________
block_1_expand_relu (ReLU)  (None, 80, 80, 96)        0            block_1_expand_BN[0][0]
__________________________________________________________________________________________________
block_1_pad (ZeroPadding2D)  (None, 81, 81, 96)     0            block_1_expand_relu[0][0]
__________________________________________________________________________________________________
block_1_depthwise (DepthwiseCon (None, 40, 40, 96)   864           block_1_pad[0][0]
__________________________________________________________________________________________________
block_1_depthwise_BN (BatchNorm (None, 40, 40, 96)   384          block_1_depthwise[0][0]
__________________________________________________________________________________________________
block_1_depthwise_relu (ReLU)   (None, 40, 40, 96)   0             block_1_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_1_project (Conv2D)     (None, 40, 40, 24)    2304           block_1_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_1_project_BN (BatchNormal  (None, 40, 40, 24)   96          block_1_project[0][0]
__________________________________________________________________________________________________
block_2_expand (Conv2D) (None, 40, 40, 144)    3456           block_1_project_BN[0][0]
__________________________________________________________________________________________________
block_2_expand_BN (BatchNormali (None, 40, 40, 144)   576          block_2_expand[0][0]
__________________________________________________________________________________________________
block_2_expand_relu (ReLU) (None, 40, 40, 144)     0           block_2_expand_BN[0][0]
__________________________________________________________________________________________________
block_2_depthwise (DepthwiseCon (None, 40, 40, 144)   1296       block_2_expand_relu[0][0]
__________________________________________________________________________________________________
block_2_depthwise_BN (BatchNorm (None, 40, 40, 144)   576      block_2_depthwise[0][0]
__________________________________________________________________________________________________
block_2_depthwise_relu (ReLU) (None, 40, 40, 144)    0        block_2_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_2_project (Conv2D)   (None, 40, 40, 24)      3456      block_2_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_2_project_BN (BatchNormal (None, 40, 40, 24)    96          block_2_project[0][0]
__________________________________________________________________________________________________
block_2_add (Add)        (None, 40, 40, 24)         0         block_1_project_BN[0][0]
block_2_project_BN[0][0]
__________________________________________________________________________________________________
block_3_expand (Conv2D)      (None, 40, 40, 144)     3456         block_2_add[0][0]
__________________________________________________________________________________________________
block_3_expand_BN (BatchNormali (None, 40, 40, 144)    576       block_3_expand[0][0]
__________________________________________________________________________________________________
block_3_expand_relu (ReLU) (None, 40, 40, 144)        0       block_3_expand_BN[0][0]
__________________________________________________________________________________________________
block_3_pad (ZeroPadding2D) (None, 41, 41, 144)   0          block_3_expand_relu[0][0]
__________________________________________________________________________________________________
block_3_depthwise (DepthwiseCon (None, 20, 20, 144)  1296         block_3_pad[0][0]
__________________________________________________________________________________________________
block_3_depthwise_BN (BatchNorm (None, 20, 20, 144)   576    block_3_depthwise[0][0]
__________________________________________________________________________________________________
block_3_depthwise_relu (ReLU)   (None, 20, 20, 144)   0         block_3_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_3_project (Conv2D)   (None, 20, 20, 32)      4608          block_3_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_3_project_BN (BatchNormal  (None, 20, 20, 32)  128      block_3_project[0][0]
__________________________________________________________________________________________________
block_4_expand (Conv2D)   (None, 20, 20, 192)     6144         block_3_project_BN[0][0]
__________________________________________________________________________________________________
block_4_expand_BN (BatchNormali (None, 20, 20, 192)   768       block_4_expand[0][0]
__________________________________________________________________________________________________
block_4_expand_relu (ReLU)   (None, 20, 20, 192)    0        block_4_expand_BN[0][0]
__________________________________________________________________________________________________
block_4_depthwise (DepthwiseCon (None, 20, 20, 192)   1728       block_4_expand_relu[0][0]
__________________________________________________________________________________________________
block_4_depthwise_BN (BatchNorm   (None, 20, 20, 192)    768       block_4_depthwise[0][0]
__________________________________________________________________________________________________
block_4_depthwise_relu (ReLU)   (None, 20, 20, 192)     0         block_4_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_4_project (Conv2D)   (None, 20, 20, 32)        6144      block_4_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_4_project_BN (BatchNormal  (None, 20, 20, 32)   128        block_4_project[0][0]
__________________________________________________________________________________________________
block_4_add (Add)         (None, 20, 20, 32)       0        block_3_project_BN[0][0]
block_4_project_BN[0][0]
__________________________________________________________________________________________________
block_5_expand (Conv2D)   (None, 20, 20, 192)      6144            block_4_add[0][0]
__________________________________________________________________________________________________
block_5_expand_BN (BatchNormali (None, 20, 20, 192)   768         block_5_expand[0][0]
__________________________________________________________________________________________________
block_5_expand_relu (ReLU) (None, 20, 20, 192)          0        block_5_expand_BN[0][0]
__________________________________________________________________________________________________
block_5_depthwise (DepthwiseCon  (None, 20, 20, 192)   1728      block_5_expand_relu[0][0]
__________________________________________________________________________________________________
block_5_depthwise_BN (BatchNorm (None, 20, 20, 192)   768       block_5_depthwise[0][0]
__________________________________________________________________________________________________
block_5_depthwise_relu (ReLU)   (None, 20, 20, 192)      0       block_5_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_5_project (Conv2D)   (None, 20, 20, 32)         6144    block_5_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_5_project_BN (BatchNormal  (None, 20, 20, 32)   128       block_5_project[0][0]
__________________________________________________________________________________________________
block_5_add (Add)           (None, 20, 20, 32)      0       block_4_add[0][0]
block_5_project_BN[0][0]
__________________________________________________________________________________________________
block_6_expand (Conv2D)     (None, 20, 20, 192)     6144          block_5_add[0][0]
__________________________________________________________________________________________________
block_6_expand_BN (BatchNormali (None, 20, 20, 192)   768     block_6_expand[0][0]
__________________________________________________________________________________________________
block_6_expand_relu (ReLU)   (None, 20, 20, 192)    0      block_6_expand_BN[0][0]
__________________________________________________________________________________________________
block_6_pad (ZeroPadding2D)  (None, 21, 21, 192)   0       block_6_expand_relu[0][0]
__________________________________________________________________________________________________
block_6_depthwise (DepthwiseCon (None, 10, 10, 192)   1728       block_6_pad[0][0]
__________________________________________________________________________________________________
block_6_depthwise_BN (BatchNorm (None, 10, 10, 192)   768         block_6_depthwise[0][0]
__________________________________________________________________________________________________
block_6_depthwise_relu (ReLU) (None, 10, 10, 192)   0    block_6_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_6_project (Conv2D) (None, 10, 10, 64)  12288         block_6_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_6_project_BN (BatchNormal (None, 10, 10, 64)   256        block_6_project[0][0]
__________________________________________________________________________________________________
block_7_expand (Conv2D) (None, 10, 10, 384)        24576        block_6_project_BN[0][0]
__________________________________________________________________________________________________
block_7_expand_BN (BatchNormali (None, 10, 10, 384)  1536        block_7_expand[0][0]
__________________________________________________________________________________________________
block_7_expand_relu (ReLU) (None, 10, 10, 384)      0         block_7_expand_BN[0][0]
__________________________________________________________________________________________________
block_7_depthwise (DepthwiseCon (None, 10, 10, 384)  3456       block_7_expand_relu[0][0]
__________________________________________________________________________________________________
block_7_depthwise_BN (BatchNorm (None, 10, 10, 384)  1536          block_7_depthwise[0][0]
__________________________________________________________________________________________________
block_7_depthwise_relu (ReLU) (None, 10, 10, 384)    0            block_7_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_7_project (Conv2D) (None, 10, 10, 64)     24576          block_7_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_7_project_BN (BatchNormal (None, 10, 10, 64)   256            block_7_project[0][0]
__________________________________________________________________________________________________
block_7_add (Add) (None, 10, 10, 64)          0               block_6_project_BN[0][0]
block_7_project_BN[0][0]
__________________________________________________________________________________________________
block_8_expand (Conv2D) (None, 10, 10, 384) 24576 block_7_add[0][0]
__________________________________________________________________________________________________
block_8_expand_BN (BatchNormali (None, 10, 10, 384) 1536 block_8_expand[0][0]
__________________________________________________________________________________________________
block_8_expand_relu (ReLU) (None, 10, 10, 384) 0 block_8_expand_BN[0][0]
__________________________________________________________________________________________________
block_8_depthwise (DepthwiseCon (None, 10, 10, 384) 3456 block_8_expand_relu[0][0]
__________________________________________________________________________________________________
block_8_depthwise_BN (BatchNorm (None, 10, 10, 384) 1536 block_8_depthwise[0][0]
__________________________________________________________________________________________________
block_8_depthwise_relu (ReLU) (None, 10, 10, 384) 0 block_8_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_8_project (Conv2D) (None, 10, 10, 64) 24576 block_8_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_8_project_BN (BatchNormal (None, 10, 10, 64) 256 block_8_project[0][0]
__________________________________________________________________________________________________
block_8_add (Add) (None, 10, 10, 64) 0 block_7_add[0][0]
block_8_project_BN[0][0]
__________________________________________________________________________________________________
block_9_expand (Conv2D) (None, 10, 10, 384) 24576 block_8_add[0][0]
__________________________________________________________________________________________________
block_9_expand_BN (BatchNormali (None, 10, 10, 384) 1536 block_9_expand[0][0]
__________________________________________________________________________________________________
block_9_expand_relu (ReLU) (None, 10, 10, 384) 0 block_9_expand_BN[0][0]
__________________________________________________________________________________________________
block_9_depthwise (DepthwiseCon (None, 10, 10, 384) 3456 block_9_expand_relu[0][0]
__________________________________________________________________________________________________
block_9_depthwise_BN (BatchNorm (None, 10, 10, 384) 1536 block_9_depthwise[0][0]
__________________________________________________________________________________________________
block_9_depthwise_relu (ReLU) (None, 10, 10, 384) 0 block_9_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_9_project (Conv2D) (None, 10, 10, 64) 24576 block_9_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_9_project_BN (BatchNormal (None, 10, 10, 64) 256 block_9_project[0][0]
__________________________________________________________________________________________________
block_9_add (Add) (None, 10, 10, 64) 0 block_8_add[0][0]
block_9_project_BN[0][0]
__________________________________________________________________________________________________
block_10_expand (Conv2D) (None, 10, 10, 384) 24576 block_9_add[0][0]
__________________________________________________________________________________________________
block_10_expand_BN (BatchNormal (None, 10, 10, 384) 1536 block_10_expand[0][0]
__________________________________________________________________________________________________
block_10_expand_relu (ReLU) (None, 10, 10, 384) 0 block_10_expand_BN[0][0]
__________________________________________________________________________________________________
block_10_depthwise (DepthwiseCo (None, 10, 10, 384) 3456 block_10_expand_relu[0][0]
__________________________________________________________________________________________________
block_10_depthwise_BN (BatchNor (None, 10, 10, 384) 1536 block_10_depthwise[0][0]
__________________________________________________________________________________________________
block_10_depthwise_relu (ReLU) (None, 10, 10, 384) 0 block_10_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_10_project (Conv2D) (None, 10, 10, 96) 36864 block_10_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_10_project_BN (BatchNorma (None, 10, 10, 96) 384 block_10_project[0][0]
__________________________________________________________________________________________________
block_11_expand (Conv2D) (None, 10, 10, 576) 55296 block_10_project_BN[0][0]
__________________________________________________________________________________________________
block_11_expand_BN (BatchNormal (None, 10, 10, 576) 2304 block_11_expand[0][0]
__________________________________________________________________________________________________
block_11_expand_relu (ReLU) (None, 10, 10, 576) 0 block_11_expand_BN[0][0]
__________________________________________________________________________________________________
block_11_depthwise (DepthwiseCo (None, 10, 10, 576) 5184 block_11_expand_relu[0][0]
__________________________________________________________________________________________________
block_11_depthwise_BN (BatchNor (None, 10, 10, 576) 2304 block_11_depthwise[0][0]
__________________________________________________________________________________________________
block_11_depthwise_relu (ReLU) (None, 10, 10, 576) 0 block_11_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_11_project (Conv2D) (None, 10, 10, 96) 55296 block_11_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_11_project_BN (BatchNorma (None, 10, 10, 96) 384 block_11_project[0][0]
__________________________________________________________________________________________________
block_11_add (Add) (None, 10, 10, 96) 0 block_10_project_BN[0][0]
block_11_project_BN[0][0]
__________________________________________________________________________________________________
block_12_expand (Conv2D) (None, 10, 10, 576) 55296 block_11_add[0][0]
__________________________________________________________________________________________________
block_12_expand_BN (BatchNormal (None, 10, 10, 576) 2304 block_12_expand[0][0]
__________________________________________________________________________________________________
block_12_expand_relu (ReLU) (None, 10, 10, 576) 0 block_12_expand_BN[0][0]
__________________________________________________________________________________________________
block_12_depthwise (DepthwiseCo (None, 10, 10, 576) 5184 block_12_expand_relu[0][0]
__________________________________________________________________________________________________
block_12_depthwise_BN (BatchNor (None, 10, 10, 576) 2304 block_12_depthwise[0][0]
__________________________________________________________________________________________________
block_12_depthwise_relu (ReLU) (None, 10, 10, 576) 0 block_12_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_12_project (Conv2D) (None, 10, 10, 96) 55296 block_12_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_12_project_BN (BatchNorma (None, 10, 10, 96) 384 block_12_project[0][0]
__________________________________________________________________________________________________
block_12_add (Add) (None, 10, 10, 96) 0 block_11_add[0][0]
block_12_project_BN[0][0]
__________________________________________________________________________________________________
block_13_expand (Conv2D) (None, 10, 10, 576) 55296 block_12_add[0][0]
__________________________________________________________________________________________________
block_13_expand_BN (BatchNormal (None, 10, 10, 576) 2304 block_13_expand[0][0]
__________________________________________________________________________________________________
block_13_expand_relu (ReLU) (None, 10, 10, 576) 0 block_13_expand_BN[0][0]
__________________________________________________________________________________________________
block_13_pad (ZeroPadding2D) (None, 11, 11, 576) 0 block_13_expand_relu[0][0]
__________________________________________________________________________________________________
block_13_depthwise (DepthwiseCo (None, 5, 5, 576)   5184       block_13_pad[0][0]
__________________________________________________________________________________________________
block_13_depthwise_BN (BatchNor (None, 5, 5, 576)   2304     block_13_depthwise[0][0]
__________________________________________________________________________________________________
block_13_depthwise_relu (ReLU) (None, 5, 5, 576)   0        block_13_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_13_project (Conv2D) (None, 5, 5, 160)        92160   block_13_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_13_project_BN (BatchNorma (None, 5, 5, 160)   640      block_13_project[0][0]
__________________________________________________________________________________________________
block_14_expand (Conv2D) (None, 5, 5, 960)     153600       block_13_project_BN[0][0]
__________________________________________________________________________________________________
block_14_expand_BN (BatchNormal (None, 5, 5, 960)   3840     block_14_expand[0][0]
__________________________________________________________________________________________________
block_14_expand_relu (ReLU) (None, 5, 5, 960)    0          block_14_expand_BN[0][0]
__________________________________________________________________________________________________
block_14_depthwise (DepthwiseCo (None, 5, 5, 960)  8640    block_14_expand_relu[0][0]
__________________________________________________________________________________________________
block_14_depthwise_BN (BatchNor (None, 5, 5, 960)  3840    block_14_depthwise[0][0]
__________________________________________________________________________________________________
block_14_depthwise_relu (ReLU) (None, 5, 5, 960)   0          block_14_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_14_project (Conv2D) (None, 5, 5, 160)     153600       block_14_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_14_project_BN (BatchNorma (None, 5, 5, 160)   640    block_14_project[0][0]
__________________________________________________________________________________________________
block_14_add (Add) (None, 5, 5, 160)         0             block_13_project_BN[0][0]
block_14_project_BN[0][0]
__________________________________________________________________________________________________
block_15_expand (Conv2D) (None, 5, 5, 960)     153600        block_14_add[0][0]
__________________________________________________________________________________________________
block_15_expand_BN (BatchNormal (None, 5, 5, 960)   3840      block_15_expand[0][0]
__________________________________________________________________________________________________
block_15_expand_relu (ReLU) (None, 5, 5, 960)   0       block_15_expand_BN[0][0]
__________________________________________________________________________________________________
block_15_depthwise (DepthwiseCo (None, 5, 5, 960)   8640      block_15_expand_relu[0][0]
__________________________________________________________________________________________________
block_15_depthwise_BN (BatchNor (None, 5, 5, 960)   3840      block_15_depthwise[0][0]
__________________________________________________________________________________________________
block_15_depthwise_relu (ReLU) (None, 5, 5, 960)    0       block_15_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_15_project (Conv2D) (None, 5, 5, 160)    153600     block_15_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_15_project_BN (BatchNorma (None, 5, 5, 160)   640      block_15_project[0][0]
__________________________________________________________________________________________________
block_15_add (Add) (None, 5, 5, 160) 0 block_14_add[0][0]
block_15_project_BN[0][0]
__________________________________________________________________________________________________
block_16_expand (Conv2D) (None, 5, 5, 960)   153600     block_15_add[0][0]
__________________________________________________________________________________________________
block_16_expand_BN (BatchNormal (None, 5, 5, 960)   3840     block_16_expand[0][0]
__________________________________________________________________________________________________
block_16_expand_relu (ReLU) (None, 5, 5, 960)    0      block_16_expand_BN[0][0]
__________________________________________________________________________________________________
block_16_depthwise (DepthwiseCo (None, 5, 5, 960)   8640       block_16_expand_relu[0][0]
__________________________________________________________________________________________________
block_16_depthwise_BN (BatchNor (None, 5, 5, 960)   3840     block_16_depthwise[0][0]
__________________________________________________________________________________________________
block_16_depthwise_relu (ReLU) (None, 5, 5, 960)    0   block_16_depthwise_BN[0][0]
__________________________________________________________________________________________________
block_16_project (Conv2D) (None, 5, 5, 320)         307200        block_16_depthwise_relu[0][0]
__________________________________________________________________________________________________
block_16_project_BN (BatchNorma (None, 5, 5, 320)         1280        block_16_project[0][0]
__________________________________________________________________________________________________
Conv_1 (Conv2D) (None, 5, 5, 1280)           409600           block_16_project_BN[0][0]
__________________________________________________________________________________________________
Conv_1_bn (BatchNormalization) (None, 5, 5, 1280)      5120          Conv_1[0][0]
__________________________________________________________________________________________________
out_relu (ReLU)        (None, 5, 5, 1280)        0            Conv_1_bn[0][0]
==================================================================================================
Total params: 2,257,984
Trainable params: 0
Non-trainable params: 2,257,984
_________________________________________________________________________

解释

从上一步创建的卷积基被冻结并用作特征提取器。
在其顶部添加一个分类器以训练顶层分类器。
冻结是通过设置`layer.trainable = False`来完成的。
此步骤避免了在训练期间更新层中的权重。
MobileNet V2有很多层，因此将模型的整个`trainable`标志设置为`False`将冻结所有层。
当`layer.trainable = False`时，BatchNormalization层以推理模式运行，不会更新均值和方差统计信息。
当模型解冻时，它包含BatchNormalization层以进行微调。
这可以通过在调用基本模型时传递`training = False`来完成。
否则，应用于不可训练权重的更新将破坏模型已学习的内容。

AmitDiwan

更新于：2021年2月25日

181 次浏览

启动你的职业生涯

完成课程获得认证

开始学习