【深度学习】VGG16“迁移学习”模型实现狗的种类识别（记录笔记）

水深无声 2022-11-01 13:55 343阅读 0赞

## 〇、内容说明 ##

接上一篇博客：  
[【深度学习】Keras自建神经网络模型实现133种狗的种类识别（记录笔记）][Keras_133]

## 一、提取数据的瓶颈特征 ##

bottleneck_features = np.load('E:/dog/DogVGG16Data.npz')
    train_VGG16 = bottleneck_features['train']
    valid_VGG16 = bottleneck_features['valid']
    test_VGG16 = bottleneck_features['test']

这里的瓶颈特征bottleneck\_features是怎么获取的？可参考[keras中文文档（VGG16）][keras_VGG16]

from keras.applications.vgg16 import VGG16
    from keras.preprocessing import image
    from keras.applications.vgg16 import preprocess_input
    import numpy as np
    
    model = VGG16(weights='imagenet', include_top=False)
    
    img_path = 'elephant.jpg'
    img = image.load_img(img_path, target_size=(224, 224))
    x = image.img_to_array(img)
    x = np.expand_dims(x, axis=0)
    x = preprocess_input(x)
    
    features = model.predict(x)
    np.save('features',features)

## 二、给VGG16安装一个我们想要的顶层 ##

“该模型使用预训练的 VGG-16 模型作为固定的图像特征提取器，其中 VGG-16 最后一层卷积层的输出被直接输入到我们的模型。我们只需要添加一个全局平均池化层以及一个全连接层，其中全连接层使用 softmax 激活函数，对每一个狗的种类都包含一个节点。”

from keras.layers import Conv2D, MaxPooling2D, GlobalAveragePooling2D
    from keras.layers import Dropout, Flatten, Dense
    from keras.models import Sequential
    VGG16_model = Sequential()
    VGG16_model.add(GlobalAveragePooling2D(input_shape=train_VGG16.shape[1:]))
    VGG16_model.add(Dense(133, activation='softmax'))
    
    VGG16_model.summary()

（512+1）\*133=68229  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70]

## 三、编译训练 ##

VGG16_model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])
    
    
    from keras.callbacks import ModelCheckpoint  
    checkpointer = ModelCheckpoint(filepath='weights.best.VGG16.hdf5', 
                                   verbose=1, save_best_only=True)
    
    VGG16_model.fit(train_VGG16, train_targets, 
              validation_data=(valid_VGG16, valid_targets),
              epochs=10, batch_size=20, callbacks=[checkpointer], verbose=1)

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 1]

## 四、测试集预测 ##

# 获取测试数据集中每一个图像所预测的狗品种的index
    VGG16_predictions = [np.argmax(VGG16_model.predict(np.expand_dims(feature, axis=0))) for feature in test_VGG16]
    
    # 报告测试准确率
    test_accuracy = 100*np.sum(np.array(VGG16_predictions)==np.argmax(test_targets, axis=1))/len(VGG16_predictions)
    print('Test accuracy: %.4f%%' % test_accuracy)

Test accuracy: 76.0766%

我直接好家伙，也是，基本上就是拿着人家训练好imagenet1000的数据集出的结果，而且这里狗狗的数据集和ISLVRC (ImageNet Large Scale Visual Recognition Challenge)比赛中狗的图片非常接近，这应该是最简易版的迁移学习吧~

## 五、预测具体狗的品种 ##

先得到预测图片的bottleneck\_features，再进行预测

#获取extract_bottleneck_features.py文件中的所有内容
    from extract_bottleneck_features import *  
    
    def VGG16_predict_breed(img_path):
        # 提取bottleneck特征
        #path_to_tensor见上篇博客
        bottleneck_feature = extract_VGG16(path_to_tensor(img_path))
        # 获取预测向量
        predicted_vector = VGG16_model.predict(bottleneck_feature)
        # 返回此模型预测的狗的品种
        return dog_names[np.argmax(predicted_vector)]
    
    print(VGG16_predict_breed(test_files[1]))

extract\_bottleneck\_features.py文件：

#Keras中常见的Application应用：
    
    def extract_VGG16(tensor):
    	from keras.applications.vgg16 import VGG16, preprocess_input
    	return VGG16(weights='imagenet', include_top=False).predict(preprocess_input(tensor))
    
    def extract_VGG19(tensor):
    	from keras.applications.vgg19 import VGG19, preprocess_input
    	return VGG19(weights='imagenet', include_top=False).predict(preprocess_input(tensor))
    
    def extract_Resnet50(tensor):
    	from keras.applications.resnet50 import ResNet50, preprocess_input
    	return ResNet50(weights='imagenet', include_top=False).predict(preprocess_input(tensor))
    
    def extract_Xception(tensor):
    	from keras.applications.xception import Xception, preprocess_input
    	return Xception(weights='imagenet', include_top=False).predict(preprocess_input(tensor))
    
    def extract_InceptionV3(tensor):
    	from keras.applications.inception_v3 import InceptionV3, preprocess_input
    	return InceptionV3(weights='imagenet', include_top=False).predict(preprocess_input(tensor))

## 六、获取bottleneck特征 ##

理论是这么获取，实际有概率会死机。。。

def extract_VGG16(tensor):
        from keras.applications.vgg16 import VGG16, preprocess_input
        return VGG16(weights='imagenet', include_top=False).predict(preprocess_input(tensor))
    
    
    bottleneck_feature = extract_VGG16(test_tensors)
    
    #保存bottleneck_feature数据
     np.save('bottleneck_feature_vgg16_test',bottleneck_feature)

1、文件名：test\_files、标签:test\_targets、数据:test\_tesnors  
![在这里插入图片描述][20210224180127499.png]  
2、经过预训练的VGG16获取到的瓶颈特征  
![在这里插入图片描述][20210224180647290.png]  
3、保存为.npy文件  
![在这里插入图片描述][20210224181003277.png]

4、加载.npy文件

aload =np.load('bottleneck_feature_vgg16_test.npy')

## 七、VGG16模型 ##

keras.applications.vgg16.VGG16(include_top=True, weights='imagenet',
                                    input_tensor=None, input_shape=None,
                                    pooling=None,
                                    classes=1000)

![include\_top：是否保留顶层的3个全连接网络weights：None代表随机初始化，即不加载预训练权重。'imagenet'代表加载预训练权重input\_tensor：可填入Keras tensor作为模型的图像输出tensorinput\_shape：可选，仅当include\_top=False有效，应为长为3的tuple，指明输入图片的shape，图片的宽高必须大于48，如(200,200,3)][include_top_3_weights_None_imagenet_input_tensor_Keras tensor_tensorinput_shape_include_top_False_3_tuple_shape_48_200_200_3]

[Keras_133]: https://blog.csdn.net/qq_41856733/article/details/114025959
[keras_VGG16]: https://keras-cn.readthedocs.io/en/latest/other/application/#resnet50
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70]: /images/20221024/f3df1ea05e844383901ae51454a16e3d.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxODU2NzMz_size_16_color_FFFFFF_t_70 1]: /images/20221024/f7bb0ba1e5914feaa85d0f8f80342d2b.png
[20210224180127499.png]: /images/20221024/89fa5fe6fba04d9cba1cd065cb8f2096.png
[20210224180647290.png]: /images/20221024/aa0365ec7d0f43beaf3d62bcdd4be15d.png
[20210224181003277.png]: /images/20221024/67c968ffbc2a4be0972066d37387b11a.png
[include_top_3_weights_None_imagenet_input_tensor_Keras tensor_tensorinput_shape_include_top_False_3_tuple_shape_48_200_200_3]: /images/20221024/ff3424b4316142c080766b15c3275a6d.png