【深度学习图像识别课程】keras实现CNN系列：（5）应用数据增强进行CIFAR10分类

一、图像增强简介

如何保证图像的标度不变性（大小）、平移不变性（位置）、旋转不变性（角度）？

 

最大池化层：保证平移不变性

增强：训练集扩展，对训练图片进行随机旋转或者平移。保证平移不变性，旋转不变性；避免过拟合

keras文档：https://keras.io/preprocessing/image/

另外的参考：https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html

                   https://machinelearningmastery.com/image-augmentation-deep-learning-keras/

 

二、CIFAR10图像库增强实战

跟增强前的CNN相比，在第4个步骤“切分数据集”之后加了：创建图像增强产生器，可视化增强图像。并且在训练模型时，使用跟增强匹配的model.fit_generator，其他均跟之前一致。

1 加载CIFAR10数据库

import keras
from keras.datasets import cifar10

(x_train, y_train),(x_test, y_test) = cifar10.load_data()
print(x_train.shape)
print(x_test.shape)

2 可视化前36幅图像

import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

fig = plt.figure(figsize=(20,5))
for i in range(36):
    ax = fig.add_subplot(3, 12, i + 1, xticks=[], yticks=[])
    ax.imshow(np.squeeze(x_train[i]))

3 归一化

x_train = x_train.astype('float32')/255
x_test = x_test.astype('float32')/255

4 切分训练集、验证集、测试集

from keras.utils import np_utils
num_classes = len(np.unique(y_train))
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

(x_train, x_valid) = x_train[5000:], x_train[:5000]
(y_train, y_valid) = y_train[5000:], y_train[:5000]

print('x_train shape:', x_train.shape)

print(x_train.shape[0], 'train examples')
print(x_valid.shape[0], 'valid examples')
print(x_test.shape[0], 'test examples')

 

5 创建和配置图像增强产生器

from keras.preprocessing.image import ImageDataGenerator

datagen_train = ImageDataGenerator(
width_shift_range = 0.1,
height_shift_range = 0.1,
horizontal_flip = True)

datagen_train.fit(x_train)

 

 

keras.preprocessing.image.ImageDataGenerator(
featurewise_center=False,
samplewise_center=False,
featurewise_std_normalization=False,
samplewise_std_normalization=False,
zca_whitening=False,
zca_epsilon=1e-06,
rotation_range=0.0,           随机旋转角度范围
width_shift_range=0.0,        宽度移动范围
height_shift_range=0.0,       高度移动范围
brightness_range=None,        亮度范围
shear_range=0.0,              剪切范围
zoom_range=0.0,               缩放方位
channel_shift_range=0.0,      通道转换范围
fill_mode='nearest',          填充模式(4种，constant, nearest, wrap, reflection)
cval=0.0,                     当填充模式为constant时，填充的值
horizontal_flip=False,        水平翻转
vertical_flip=False,          垂直翻转
rescale=None,                 数据缩放
preprocessing_function=None,  图像缩放、增强后使用
data_format=None,             图像集格式，(samples,height,width,channels)还是channels在samples后
validation_split=0.0)         数据集用来做验证集的比例

 

6 可视化原始和增强后的图像

 

import matplotlib.pyplot as plt

# take subset of training data
x_train_subset = x_train[:12]

# visualize subset of training data
fig = plt.figure(figsize=(20,2))
for i in range(0, len(x_train_subset)):
    ax = fig.add_subplot(1, 12, i+1)
    ax.imshow(x_train_subset[i])
fig.suptitle('Subset of Original Training Images', fontsize=20)
plt.show()

# visualize augmented images
fig = plt.figure(figsize=(20,2))
for x_batch in datagen_train.flow(x_train_subset, batch_size=12):
    for i in range(0, 12):
        ax = fig.add_subplot(1, 12, i+1)
        ax.imshow(x_batch[i])
    fig.suptitle('Augmented Images', fontsize=20)
    plt.show()
    break;

 

7 定义模型

 

from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout

model = Sequential()
model.add(Conv2D(filters=16, kernel_size=2, padding='same', activation='relu', input_shape=(32,32,3)))
model.add(MaxPooling2D(pool_size=2))
model.add(Conv2D(filters=32, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Conv2D(filters=64, kernel_size=2, padding='same', activation='relu'))
model.add(MaxPooling2D(pool_size=2))
model.add(Dropout(0.3))
model.add(Flatten())
model.add(Dense(500, activation='relu'))
model.add(Dropout(0.4))
model.add(Dense(num_classes, activation='softmax'))

model.summary()

模型跟增强前比，不变。

8 编译模型

model.compile(loss='categorical_crossentropy', optimizer='rmsprop', metrics=['accuracy'])

9 训练模型

from keras.callbacks import ModelCheckpoint

batch_size = 32
checkpoint = ModelCheckpoint(filepath='MLP.weights.best.hdf5', verbose=1, save_best_only=True)
model.fit_generator(datagen_train.flow(x_train, y_train, batch_size=32),
                    steps_per_epoch=x_train.shape[0] // batch_size,
                    epochs = 100,
                    verbose=2,
                    callbacks=[checkpoint],
                    validation_data=(x_valid, y_valid),
                    validation_steps=x_valid.shape[0] // batch_size)

参数说明：

flow(x, y=None, batch_size=32, shuffle=True, seed=None, save_to_dir=None, save_prefix='', save_format='png', subset=None)

• x: 数据集
• y: 标签
• batch_size: 
• shuffle: 
• seed: 
• save_to_dir: 可以设置为None或者字符串str。如果设置为str，则可以保存产生的增强图像。
• save_prefix: 保存增强图像名的前缀。如果变量save_to_dir设置为路径，这里才有效。
• save_format: 增强图像保存格式，可以为png或者jpeg，默认png。同样，.变量save_to_dir设置为路径，这里才有效。
• subset: 子集("training" or "validation")。当ImageDataGenerator 中 validation_split 设置才有效。

fit_generator(self, generator, steps_per_epoch=None, epochs=1, verbose=1, callbacks=None, validation_data=None, validation_steps=None, class_weight=None, max_queue_size=10, workers=1, use_multiprocessing=False, shuffle=True, initial_epoch=0)

• generator:
• steps_per_epoch = x_train.shape[0]/batch_size。确保模型在每个epoch中看 x_train.shape[0]个增强图片
• epochs:
• verbose: 0, 1, or 2
• callbacks:
• validation_data:
• validation_steps:
• class_weight:
• max_queue_size:
• workers:
• use_multiprocessing:
• shuffle:
• initial_epoch: 初始epoch

 

10 加载在验证集上分类正确率最高的一组模型参数

model.load_weights('MLP.weights.best.hdf5')

11 测试集上计算分类正确率

score = model.evaluate(x_test, y_test, verbose=0)
print('\n', 'Test accuracy:', score[1])

我只训练了10个epoch，理论上这里的正确率应该是要比没有增强前的高。