Keras —— 基于InceptionV3模型（不含全连接层）的迁移学习应用

一、ImageDataGenerator

def image_preprocess():
    # 　训练集的图片生成器，通过参数的设置进行数据扩增
    train_datagen = ImageDataGenerator(
        preprocessing_function=preprocess_input,
        rotation_range=30,
        width_shift_range=0.2,
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True
    )
    #   验证集的图片生成器，不进行数据扩增，只进行数据预处理
    val_datagen = ImageDataGenerator(
        preprocessing_function=preprocess_input,
    )
    # 训练数据与测试数据
    train_generator = train_datagen.flow_from_directory(
        train_dir,
        target_size=(IM_WIDTH, IM_HEIGHT),
        batch_size=batch_size, class_mode='categorical')

    validation_generator = val_datagen.flow_from_directory(
        val_dir,
        target_size=(IM_WIDTH, IM_HEIGHT),
        batch_size=batch_size, class_mode='categorical')
    return train_generator, validation_generator

二、加载InceptionV3模型（不含全连接层）

使用带有预训练权重的InceptionV3模型，但不包括顶层分类器（顶层分类器即全连接层。）

base_model = InceptionV3(weights='imagenet', include_top=False)

三、添加新的顶层分类器

def add_new_last_layer(base_model, nb_classes):
    x = base_model.output
    x = GlobalAveragePooling2D()(x)
    x = Dense(FC_SIZE, activation='relu')(x)
    predictions = Dense(nb_classes, activation='softmax')(x)
    model = Model(input=base_model.input, output=predictions)
    return model

四、训练顶层分类器

冻结base_model所有层，这样就可以正确获得bottleneck特征

def setup_to_transfer_learn(model, base_model):
    for layer in base_model.layers:
        layer.trainable = False
    model.compile(optimizer='rmsprop', loss='categorical_crossentropy', metrics=['accuracy'])

setup_to_transfer_learn(model, base_model)
history_tl = model.fit_generator(
        train_generator,
        epochs=nb_epoch,
        steps_per_epoch=nb_train_samples // batch_size,
        validation_data=validation_generator,
        validation_steps=nb_val_samples // batch_size,
        class_weight='auto')

五、对顶层分类器进行fine_tuning

冻结部分层，对顶层分类器进行Fine-tune

Fine-tune以一个预训练好的网络为基础，在新的数据集上重新训练一小部分权重。fine-tune应该在很低的学习率下进行，通常使用SGD优化

def setup_to_finetune(model):
    for layer in model.layers[:NB_IV3_LAYERS_TO_FREEZE]:
        layer.trainable = False
    for layer in model.layers[NB_IV3_LAYERS_TO_FREEZE:]:
        layer.trainable = True
    model.compile(optimizer=SGD(lr=0.0001, momentum=0.9), loss='categorical_crossentropy', metrics=['accuracy'])

setup_to_finetune(model)  # 冻结model的部分层
history_ft = model.fit_generator(
        train_generator,
        steps_per_epoch=nb_train_samples // batch_size,
        epochs=nb_epoch,
        validation_data=validation_generator,
        validation_steps=nb_val_samples // batch_size,
        class_weight='auto')

---

源码地址：

https://github.com/Zheng-Wenkai/Keras_Demo