بازشناسی تصویر با Keras و شبکه های عصبی پیچشی — راهنمای کاربردی

1from keras.layers import Conv2D
2from keras.layers import Dense
3from keras.layers import Flatten
4from keras.preprocessing import image
5from keras.layers import MaxPooling2D
6from keras.models import Sequential
7import numpy as np

1classifier = Sequential()
2classifier.add(Conv2D(32, (3, 3), input_shape=(64, 64, 3),
3               activation='relu'))
4classifier.add(MaxPooling2D(pool_size=(2, 2)))
5
6classifier.add(Conv2D(32, (3, 3), activation='relu'))
7classifier.add(MaxPooling2D(pool_size=(2, 2)))
8classifier.add(Flatten())
9
10classifier.add(Dense(units=128, activation='relu'))
11classifier.add(Dense(units=1, activation='sigmoid'))

1classifier.compile(optimizer='adam', loss='binary_crossentropy',
2                   metrics=['accuracy'])
3from keras.preprocessing.image import ImageDataGenerator
4train_imagedata = ImageDataGenerator(rescale=1. / 255, shear_range=0.2,
5        zoom_range=0.2, horizontal_flip=True)
6test_imagedata = ImageDataGenerator(rescale=1. / 255)
7training_set = \
8    train_imagedata.flow_from_directory('/home/directory/image classification/data/training_set'
9        , target_size=(64, 64), batch_size=32, class_mode='binary')
10test_set = \
11    test_imagedata.flow_from_directory('/home/directory/image classification/data/test_set'
12        , target_size=(64, 64), batch_size=32, class_mode='binary')
13history=classifier.fit_generator(training_set, steps_per_epoch=80, epochs=20,
14                         validation_data=test_set,
15                         validation_steps=80)

1import matplotlib.pyplot as plt
2print(history.history.keys())
3# Loss
4plt.plot(history.history['loss'])
5plt.plot(history.history['val_loss'])
6plt.title('model loss')
7plt.ylabel('loss')
8plt.xlabel('epoch')
9plt.legend(['loss', 'val_loss'], loc='upper left')
10plt.show()
11# Accuracy
12plt.plot(history.history['acc'])
13plt.plot(history.history['val_acc'])
14plt.title('model accuracy')
15plt.ylabel('accuracy')
16plt.xlabel('epoch')
17plt.legend(['acc', 'val_acc'], loc='upper left')
18plt.show()

1classifier.compile(optimizer='adam', loss='binary_crossentropy',
2                   metrics=['accuracy'])
3from keras.preprocessing.image import ImageDataGenerator
4train_imagedata = ImageDataGenerator(rescale=1. / 255, shear_range=0.2,
5        zoom_range=0.2, horizontal_flip=True)
6test_imagedata = ImageDataGenerator(rescale=1. / 255)
7training_set = \
8    train_imagedata.flow_from_directory('/home/directory/image classification/data/training_set'
9        , target_size=(64, 64), batch_size=32, class_mode='binary')
10test_set = \
11    test_imagedata.flow_from_directory('/home/directory/image classification/data/test_set'
12        , target_size=(64, 64), batch_size=32, class_mode='binary')
13history=classifier.fit_generator(training_set, steps_per_epoch=80, epochs=2,
14                         validation_data=test_set,
15                         validation_steps=80)

1>>> test_image = \
2...     image.load_img('/home/directory/image classification/data/7813125.jpg'
3...                    , target_size=(64, 64))
4>>> test_image = image.img_to_array(test_image)
5>>> test_image = np.expand_dims(test_image, axis=0)
6>>> result = classifier.predict(test_image)
7>>> training_set.class_indices
8{'cars': 0, 'planes': 1}
9>>> result
10array([[0.]], dtype=float32)

1>>> test_image = \
2...     image.load_img('/home/directory/image classification/data/56315795.jpg'
3...                    , target_size=(64, 64))
4>>> test_image = image.img_to_array(test_image)
5>>> test_image = np.expand_dims(test_image, axis=0)
6>>> result = classifier.predict(test_image)
7>>> training_set.class_indices
8{'cars': 0, 'planes': 1}
9>>> result
10array([[1.]], dtype=float32)