下载数据集
网上找的数据集，和项目的数据集数目不同，但是无所谓了
提取码：jzyc
划分数据
需要向整体的数据划分位训练集和测试集
下载下来的zip文件中有img文件夹，内部图片0.xx.jpeg是猫类型图片
1.xx.jpeg是狗的类型图片，总共有12000张图片，每个种类个6000张
需要将器划分为
import random
cat_dir = "E:/datasets/tmp/cat_dog/cat_dog/cat/"
dog_dir = "E:/datasets/tmp/cat_dog/cat_dog/dog/"
train_cat_dir = "E:/datasets/tmp/cat_dog/training/cat/"
train_dog_dir = "E:/datasets/tmp/cat_dog/training/dog/"
test_cat_dir = "E:/datasets/tmp/cat_dog/testing/cat/"
test_dog_dir = "E:/datasets/tmp/cat_dog/testing/dog/"
def split_data(source, train, test, split_size):
"""
根据传入的源文件夹，将总共的图片经过一定的比例放到训练和测试的文件夹中
param:
source: 要切分的所有图片的总文件夹
train: 要保存训练集图片的文件夹
test: 要保存测试集图片的文件夹
split_size: 切分的比例
return:
None
"""
files = []
for filename in os.listdir(source):
file = source + '/' + filename
files.append(filename)
train_len = int(len(files) * split_size)
test_len = len(files) - train_len
shuffled_set = random.sample(files, len(files))
train_set = shuffled_set[0 : train_len]
test_set = shuffled_set[-test_len: ]
for filename in train_set:
this_file = source + filename
destination = train + filename
copyfile(this_file, destination)
for filename in test_set:
this_file = source + filename
destination = test + filename
copyfile(this_file, destination)
print("{}文件写入成功 训练集大小{}, 测试集大小{}".format(source.split('/')[-2] ,len(os.listdir(train)), len(os.listdir(test))))
split_data(cat_dir, train_cat_dir, test_cat_dir, 0.9)
split_data(dog_dir, train_dog_dir, test_dog_dir, 0.9)
数据预处理
图片格式归一化，规定batch_size，以及class类型
train_dir = "E:/datasets/tmp/cat_dog/training/"
validation_dir = "E:/datasets/tmp/cat_dog/testing/"
train_datagen = ImageDataGenerator(rescale=1/255)
train_generator = train_datagen.flow_from_directory(train_dir,batch_size=54, class_mode='binary', target_size=(100, 100))
validation_datagen = ImageDataGenerator(rescale=1/255)
validation_generator = validation_datagen.flow_from_directory(validation_dir,batch_size=60, class_mode='binary', target_size=(100, 100))
搭建神经网络
采用Le_Net5网络
def netWork():
model = tf.keras.Sequential()
model.add(tf.keras.layers.Conv2D(32,(3, 3), input_shape=(100, 100, 3), activation='relu', name='conv-1'))
model.add(tf.keras.layers.MaxPooling2D((2, 2), name='pool-1'))
model.add(tf.keras.layers.Conv2D(64, (3, 3), activation='relu', name='conv-2'))
model.add(tf.keras.layers.MaxPooling2D((2, 2), name='pool-2'))
model.add(tf.keras.layers.Conv2D(128, (3, 3), activation='relu', name='conv-3'))
model.add(tf.keras.layers.MaxPooling2D((2, 2), name='pool-3'))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(512, activation='relu', name='fully-1'))
model.add(tf.keras.layers.Dense(1, activation='sigmoid', name='sigmoid-1'))
model.compile(optimizer=RMSprop(learning_rate=0.01), loss=tf.losses.binary_crossentropy, metrics=['accuracy'])
model.summary()
return model
fit操作
model = netWork()
model.fit(train_generator,
epochs=3,
verbose=1,
validation_data=validation_generator)
制作散点图
import matplotlib.pyplot as plt
acc = history.history['accuracy']
val_acc = history.history['val_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(len(acc))
plt.plot(epochs, acc, 'r', "Train accuracy")
plt.plot(epochs, val_acc, 'b', f"Test accuracy")
plt.figure()
plt.plot(epochs, loss, 'r', "Train loss")
plt.plot(epochs, val_loss, 'b', "Test loss")
plt.figure()
自测数据
自己添加数据到项目中，来检查模型训练的好坏
from tensorflow.keras.preprocessing import image
import numpy as np
path = './image/'
index = 1
for file in os.listdir(path):
img = image.load_img(path+file, target_size=(100, 100))
x= image.img_to_array(img)
x = np.expand_dims(x, axis=0)
images = np.vstack([x])
classes = model.predict(images, batch_size=10)
print(classes[0])
plt.subplot(1,5,index)
plt.imshow(img)
num = file.split('.')[0]
if num == '0':
plt.title('cat')
else:
plt.title('dog')
if classes[0]>0.5:
print(file + " is dog")
else:
print(file + " is cat")
index += 1
