利用CNN进行猫狗分类

竞赛介绍：Kaggle Dogs vs. Cats （https://www.kaggle.com/c/dogs-vs-cats）

要点：

1. 用kaggle API下载数据后，train文件夹下的猫狗图片须分别归入2个文件夹，即cat和dog，否则flow_from_directory会报错

2. 由于该竞赛项目已经结束，本示例没有对test文件夹下的图片进行分类，而是用train文件夹下的图片进行训练和验证

3. train文件夹下共有25000张图片，其中猫狗各有12500张

代码部分：

# 加载libraries

import os

import numpy as np

import pandas as pd

import matplotlib.pyplot as plt

import matplotlib.figure as fig

import tensorflow as tf

from tensorflow.keras.preprocessing.image import ImageDataGenerator

# 设置文件路径

dir = os.getcwd()

train_dir = os.path.join(dir, 'train')

# 显示train文件夹下的猫狗图片

fig = plt.gcf()

fig.set_size_inches(10,10)

for i in range(9):

    plt.subplot(330 + 1 + i)

    file_name = train_dir + '\\dog\\dog.' + str(i) + '.jpg'

    im = plt.imread(file_name)

    plt.imshow(im)

fig = plt.gcf()

fig.set_size_inches(10,10)

for i in range(9):

    plt.subplot(330 + 1 + i)

    file_name = train_dir + '\\cat\\cat.' + str(i) + '.jpg'

    im = plt.imread(file_name)

    plt.imshow(im)

# 定义earlystopping，若验证数据集的精度在2个epoch后不再改进，则停止model fit

monitor_val_acc = tf.keras.callbacks.EarlyStopping(monitor='val_accuracy', patience=2)

# 定义model

model = tf.keras.models.Sequential([

    tf.keras.layers.Conv2D(filters = 32, kernel_size = (3,3), activation = 'relu', input_shape = (150,150,3)),

    tf.keras.layers.MaxPooling2D(pool_size = (2,2)),

    tf.keras.layers.Conv2D(filters = 64, kernel_size = (3,3), activation = 'relu'),

    tf.keras.layers.MaxPooling2D(pool_size = (2,2)),

    tf.keras.layers.Conv2D(filters = 128, kernel_size = (3,3), activation = 'relu'),

    tf.keras.layers.MaxPooling2D(pool_size = (2,2)),

    tf.keras.layers.Conv2D(filters = 128, kernel_size = (3,3), activation = 'relu'),

    tf.keras.layers.MaxPooling2D(pool_size = (2,2)),

    tf.keras.layers.Flatten(),

    tf.keras.layers.Dense(units = 512, activation = 'relu'),

    tf.keras.layers.Dense(units = 1, activation = 'sigmoid')   

])

# 编译model

model.compile(loss = 'binary_crossentropy',optimizer = 'adam', metrics = ['accuracy'])

# 定义ImageDataGenerator，同时考虑图像增强；如需将train数据集划分为训练和验证两个子集，需在此设置validation_split

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',

                                  validation_split=0.2

                                  )

# 定义train_generator和validate_generator，classes根据label进行设置，class_mode根据应用场景设置（二分类为binary），subset根据用途分别设置为training和validation

train_generator = train_datagen.flow_from_directory(directory = train_dir,

                                                  target_size = (150,150),

                                                  classes = ['cat','dog'],

                                                    batch_size = 20,

                                                  class_mode = 'binary',

                                                  subset = 'training')

validate_generator = train_datagen.flow_from_directory(directory = train_dir,

                                                      target_size = (150,150),

                                                      classes = ['cat','dog'],

                                                      batch_size = 20,

                                                      class_mode = 'binary',

                                                      subset = 'validation')

Found 20000 images belonging to 2 classes.

Found 5000 images belonging to 2 classes.

# 训练model

history = model.fit_generator(generator = train_generator,

                            steps_per_epoch = 1000,

                            epochs = 20,

                            validation_data = validate_generator,

                            validation_steps = 250,

                              callbacks = [monitor_val_acc],

                              verbose = 2)

Epoch 1/20

1000/1000 - 795s - loss: 0.5794 - accuracy: 0.6880 - val_loss: 0.4907 - val_accuracy: 0.7618

Epoch 2/20

1000/1000 - 786s - loss: 0.4575 - accuracy: 0.7836 - val_loss: 0.3896 - val_accuracy: 0.8212

Epoch 3/20

1000/1000 - 804s - loss: 0.3608 - accuracy: 0.8391 - val_loss: 0.3579 - val_accuracy: 0.8384

Epoch 4/20

1000/1000 - 772s - loss: 0.2954 - accuracy: 0.8714 - val_loss: 0.3543 - val_accuracy: 0.8448

Epoch 5/20

1000/1000 - 765s - loss: 0.2313 - accuracy: 0.9015 - val_loss: 0.3222 - val_accuracy: 0.8662

Epoch 6/20

1000/1000 - 780s - loss: 0.1746 - accuracy: 0.9313 - val_loss: 0.3112 - val_accuracy: 0.8724

Epoch 7/20

1000/1000 - 797s - loss: 0.1204 - accuracy: 0.9523 - val_loss: 0.3935 - val_accuracy: 0.8784

Epoch 8/20

1000/1000 - 789s - loss: 0.0882 - accuracy: 0.9669 - val_loss: 0.4920 - val_accuracy: 0.8692

Epoch 9/20

1000/1000 - 800s - loss: 0.0594 - accuracy: 0.9785 - val_loss: 0.4468 - val_accuracy: 0.8770

训练数据集精度为0.9785，验证数据集精度为0.8770

# 绘制learning curves图

loss = history.history['loss']

val_loss = history.history['val_loss']

accuracy = history.history['accuracy']

val_accuracy = history.history['val_accuracy']

epoch = range(len(loss))

plt.style.use('ggplot')

plt.plot(epoch, loss, color = 'blue', label = 'training loss')

plt.plot(epoch, val_loss, color = 'red', label = 'validation loss')

plt.title('model loss', size = 20)

plt.legend()

plt.figure()

plt.plot(epoch, accuracy, color = 'blue', label = 'training accuracy')

plt.plot(epoch, val_accuracy, color = 'red', label = 'validation accuracy')

plt.title('model accuracy', size = 20)

plt.legend()