Keras图像打分（3：建立网络）

定义网络结构

batch_size = 32

def CNNmodel():
    input_layer=Input(shape=(224,224,3))
    dense_model=DenseNet121(include_top=False,weights="imagenet",input_tensor=input_layer)
    dense=dense_model(input_layer)

    #对dense_121进行全局平均池化
    top_model=layers.GlobalAveragePooling2D()(dense)
#     top_model = Dense(1024, activation='relu',kernel_regularizer=regularizers.l2(0.0001), )(t)
#     #top_model = BatchNormalization(name='bn_fc_01')(top_model)
#     top_model = layers.Dropout(0.5)(top_model)
#     top_model = Dense(512,  activation='relu',kernel_regularizer=regularizers.l2(0.0001))(top_model)
#     top_model = layers.Dropout(0.3)(top_model)
    top_model = layers.Dense(1)(top_model)
    #top_model = layers.Activation("linear")(top_model)
    
    model=Model(inputs=input_layer,outputs=top_model)
    for layer in dense_model.layers:
        layer.trainable = False
    
    return model
model = CNNmodel()
model.summary()

编译模型并训练

model.compile(optimizer=optimizers.Adam(lr=0.001,beta_1=0.9,beta_2=0.999,epsilon=1e-8),
              loss='mse', metrics=['mae'])

history = model.fit_generator(generator=train_image_generator(train_files, batch_size),
                              steps_per_epoch=len(train_files)//batch_size,
                              validation_data=val_image_generator(val_files, batch_size),
                              validation_steps=len(val_files)//batch_size,
                              epochs=30)

保存模型

model.save("DenseNet.hdf5")