利用Inception-V3训练的权重微调实现猫狗的分类，其中权重的下载在我的博客下载资源处，https://download.csdn.net/download/fanzonghao/10566634

第一种权重不改变直接用mixed7层（mixed7呆会把打印结果一放就知道了）进行特征提取，然后在拉平，连上两层神经网络

def define_model():InceptionV3_weight_path='./model_weight/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5'pre_trained_model=InceptionV3(input_shape=(150,150,3),include_top=False,#不包含全连接层weights=None)pre_trained_model.load_weights(InceptionV3_weight_path)#下面两种取其一#仅仅用其做特征提取 不需要更新权值for layer in pre_trained_model.layers:print(layer.name)layer.trainable=False#微调权值# unfreeze=False# for layer in pre_trained_model.layers:#     if unfreeze:#         layer.trainable=True#     if layer.name=='mixed6':#         unfreeze=Truelast_layer=pre_trained_model.get_layer('mixed7')print(last_layer.output_shape)last_output=last_layer.output#以下是在模型的基础上增加的x=layers.Flatten()(last_output)x=layers.Dense(1024,activation='relu')(x)x=layers.Dropout(0.2)(x)x=layers.Dense(1,activation='sigmoid')(x)model=Model(inputs=pre_trained_model.input,outputs=x)return model

第一种完全利用Inception-V3训练的权重代码

import os
import tensorflow as tf
import matplotlib.pyplot as pltfrom keras.applications.inception_v3 import InceptionV3
from keras import  layers
from keras.models import Model
from keras.optimizers import RMSprop
from keras import backend as K
from keras.preprocessing.image import ImageDataGenerator
import data_read
"""
#获得所需求的图片--进行了图像增强
"""
def data_deal_overfit():# 获取数据的路径train_dir, validation_dir, next_cat_pix, next_dog_pix = data_read.read_data()#图像增强train_datagen=ImageDataGenerator(rescale=1./255,rotation_range=40,width_shift_range=0.2,height_shift_range=0.2,shear_range=0.2,zoom_range=0.2,horizontal_flip=True,fill_mode='nearest')test_datagen=ImageDataGenerator(rescale=1./255)#从文件夹获取所需要求的图片train_generator=train_datagen.flow_from_directory(train_dir,target_size=(150,150),batch_size=20,class_mode='binary')test_generator = test_datagen.flow_from_directory(validation_dir,target_size=(150, 150),batch_size=20,class_mode='binary')return train_generator,test_generator
"""
#定义模型并加入了dropout
"""
def define_model():InceptionV3_weight_path='./model_weight/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5'pre_trained_model=InceptionV3(input_shape=(150,150,3),include_top=False,#不包含全连接层weights=None)pre_trained_model.load_weights(InceptionV3_weight_path)#下面两种取其一#仅仅用其做特征提取 不需要更新权值for layer in pre_trained_model.layers:print(layer.name)layer.trainable=False#微调权值# unfreeze=False# for layer in pre_trained_model.layers:#     if unfreeze:#         layer.trainable=True#     if layer.name=='mixed6':#         unfreeze=Truelast_layer=pre_trained_model.get_layer('mixed7')print(last_layer.output_shape)last_output=last_layer.output#以下实在模型的基础上增加的x=layers.Flatten()(last_output)x=layers.Dense(1024,activation='relu')(x)x=layers.Dropout(0.2)(x)x=layers.Dense(1,activation='sigmoid')(x)model=Model(inputs=pre_trained_model.input,outputs=x)return model"""
训练模型
"""
def train_model():model=define_model()model.compile(optimizer=RMSprop(lr=0.001), loss='binary_crossentropy', metrics=['accuracy'])train_generator, test_generator = data_deal_overfit()# verbose：日志显示，0为不在标准输出流输出日志信息，1为输出进度条记录，2为每个epoch输出一行记录# 训练模型 返回history包含各种精度和损失history = model.fit_generator(train_generator,steps_per_epoch=100,  # 2000 images=batch_szie*stepsepochs=50,validation_data=test_generator,validation_steps=50,  # 1000=20*50verbose=2)#精度acc=history.history['acc']val_acc=history.history['val_acc']#损失loss=history.history['loss']val_loss=history.history['val_loss']#epochs的数量epochs=range(len(acc))plt.plot(epochs,acc)plt.plot(epochs, val_acc)plt.title('training and validation accuracy')plt.figure()plt.plot(epochs, loss)plt.plot(epochs, val_loss)plt.title('training and validation loss')plt.show()if __name__ == '__main__':train_model()
打印结果：其中这些代表每一层的名字，直接利用mixed7的特征，（none,7,7,768）就是该层的shape, 直接拉平，添加两层神经网络进行分类。

打印结果：这是每一层的名字，mixed7层的shape是（None,7,7,768）第一种做法就是直接利用该层及之前层的权重进行训练分类的。

第二种：进行微调要不是需要对整个权重都进行重新赋值，因为前面层数学习到的特征是一些简单的特征,只是随着层数增强才更加具有针对性，故把mixed7层的卷积层权重 重新训练，代码：

unfreeze=False
for layer in pre_trained_model.layers:if unfreeze:layer.trainable=Trueif layer.name=='mixed6':unfreeze=True

也就是把我上段完整的代码注释替换一下即可。