一、Keras的应用模块

Kera的应用模块Application提供了带有预训练权重的Keras模型，这些模型可以用来进行预测、特征提取和finetune。

当前keras已有的model如图：

本次实战采用的denseNet。

denseNet源代码和各个模型的比较可参考github上keras-team的keras-applications项目：

Reference implementations of popular deep learning models

说明：如果报错No model named densenet,原因可能是因为keras版本没有更新。我直接把keras更新到了最新版

指令：pip install --upgrade keras,之后又出现了keras版本和tensorflow版本不兼容的问题，通过修改tensorflow的源代码解决了。

二、DenseNet

获取模型信息的代码

#--coding:utf-8--
#获得模型信息的代码
from keras.applications.densenet import DenseNet201,preprocess_input
from keras.layers import Dense, GlobalAveragePooling2D
from keras.models import Model

#base_model = DenseNet(weights='imagenet', include_top=False)
base_model = DenseNet201(include_top=False)

x = base_model.output
x = GlobalAveragePooling2D()(x) 
predictions = Dense(5, activation='softmax')(x)
model = Model(inputs=base_model.input, outputs=predictions)

model.summary()
print('the number of layers in this model:'+str(len(model.layers)))

输出结果很长，最后部分截图：

可以看到这个模型densenet201有709layer.

三、数据集

采用的数据集同【keras实战】用Inceptionv3实现五种花的分类

四、训练

系统：Ubuntu 16.04 LTS

GPU:GeForce GTX 1080 Ti

代码：

# --coding:utf-8--
import os
import sys
import glob
import matplotlib.pyplot as plt

from keras import __version__
from keras.applications.densenet import DenseNet201,preprocess_input

from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import SGD
from keras.callbacks import ModelCheckpoint,TensorBoard


def get_nb_files(directory):
  """Get number of files by searching directory recursively"""
  if not os.path.exists(directory):
    return 0
  cnt = 0
  for r, dirs, files in os.walk(directory):
    for dr in dirs:
      cnt += len(glob.glob(os.path.join(r, dr + "/*")))
  return cnt


# 数据准备
IM_WIDTH, IM_HEIGHT = 224, 224 #densenet指定的图片尺寸
#FC_SIZE = 1024                # 全连接层的节点个数
#NB_IV3_LAYERS_TO_FREEZE = 172  # 冻结层的数量


train_dir = '/home/pandafish/AnacondaProjects/DenseNet/dataset_flower2/train'  # 训练集数据
val_dir = '/home/pandafish/AnacondaProjects/DenseNet/dataset_flower2/validate' # 验证集数据
nb_classes= 5
nb_epoch = 30
batch_size = 32

nb_train_samples = get_nb_files(train_dir)      # 训练样本个数
nb_classes = len(glob.glob(train_dir + "/*"))  # 分类数
nb_val_samples = get_nb_files(val_dir)       #验证集样本个数
nb_epoch = int(nb_epoch)                # epoch数量
batch_size = int(batch_size)           

#　图片生成器
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
)
test_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
)

# 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 = test_datagen.flow_from_directory(
val_dir,
target_size=(IM_WIDTH, IM_HEIGHT),
batch_size=batch_size,class_mode='categorical')

# 添加新层
def add_new_last_layer(base_model, nb_classes):
  """
  添加最后的层
  输入
  base_model和分类数量
  输出
  新的keras的model
  """
  x = base_model.output
  x = GlobalAveragePooling2D()(x)
  #x = Dense(FC_SIZE, activation='relu')(x) #new FC layer, random init
  predictions = Dense(nb_classes, activation='softmax')(x) #new softmax layer
  model = Model(input=base_model.input, output=predictions)
  return model


#搭建模型
model = DenseNet201(include_top=False)
model = add_new_last_layer(model, nb_classes)
model.load_weights('model/checkpoint-02e-val_acc_0.82.hdf5')
model.compile(optimizer=SGD(lr=0.001, momentum=0.9,decay=0.0001,nesterov=True), loss='categorical_crossentropy', metrics=['accuracy'])


#更好地保存模型 Save the model after every epoch.
output_model_file = '/home/pandafish/AnacondaProjects/DenseNet/model/checkpoint-{epoch:02d}e-val_acc_{val_acc:.2f}.hdf5'
#keras.callbacks.ModelCheckpoint(filepath, monitor='val_loss', verbose=0, save_best_only=False, save_weights_only=False, mode='auto', period=1)
checkpoint = ModelCheckpoint(output_model_file, monitor='val_acc', verbose=1, save_best_only=True)

#tensorboard可视化
RUN = RUN + 1 if 'RUN' in locals() else 1   # locals() 函数会以字典类型返回当前位置的全部局部变量。
LOG_DIR = '/home/pandafish/AnacondaProjects/DenseNet/training_logs/run{}'.format(RUN)
tensorboard = TensorBoard(log_dir=LOG_DIR, write_images=True)


#开始训练
history_ft = model.fit_generator(
train_generator,
samples_per_epoch=nb_train_samples,
nb_epoch=nb_epoch,
callbacks=[tensorboard,checkpoint],
validation_data=validation_generator,
nb_val_samples=nb_val_samples)

tensorboar 可视化训练显示结果（深色线是曲线拟合的结果，浅色线是实际曲线）：

训练时我迭代到第12个epoch就停了，事实上从第4个epoch开始验证集的准确率就开始不提升了，之后就过拟合了……

四、测试

# --coding:utf-8--
# 定义层
import sys
import argparse
import numpy as np
from PIL import Image
from io import BytesIO
import matplotlib.pyplot as plt

from keras.preprocessing import image
from keras.models import load_model
from keras.applications.densenet import preprocess_input

# 狂阶图片指定尺寸
target_size = (224, 224) 

# 预测函数
# 输入：model，图片，目标尺寸
# 输出：预测predict
def predict(model, img, target_size):
  """Run model prediction on image
  Args:
    model: keras model
    img: PIL format image
    target_size: (w,h) tuple
  Returns:
    list of predicted labels and their probabilities
  """
  if img.size != target_size:
    img = img.resize(target_size)

  x = image.img_to_array(img)
  x = np.expand_dims(x, axis=0)
  x = preprocess_input(x)
  preds = model.predict(x)
  return preds[0]

# 画图函数
# 预测之后画图，这里默认是猫狗，当然可以修改label

labels = ("daisy", "dandelion","roses","sunflowers","tulips")
def plot_preds(image, preds,labels):
  """Displays image and the top-n predicted probabilities in a bar graph
  Args:
    image: PIL image
    preds: list of predicted labels and their probabilities
  """
  plt.imshow(image)
  plt.axis('off')
  plt.figure()
  plt.barh([0, 1,2,3,4], preds, alpha=0.5)
  plt.yticks([0, 1,2,3,4], labels)
  plt.xlabel('Probability')
  plt.xlim(0,1.01)
  plt.tight_layout()
  plt.show()

# 载入模型
model = load_model('model/checkpoint-08e-val_acc_0.96.hdf5')

# 本地图片
img = Image.open('rose.jpg')
preds = predict(model, img, target_size)
plot_preds(img, preds,labels)

测试结果：

【keras实战】用DenseNet实现五种花的分类

一、Keras的应用模块

二、DenseNet

三、数据集

四、训练

四、测试

猜你喜欢