Aller au contenu

Les réseaux convolutifs

Un réseau convolutionnel est un réseau de neurones. Une succession de couches de convolution et de pooling.

Convolution

  1. Nombre de filtres K
  2. Taille des filtres F
  3. Le pas S
  4. Le zero-padding

Pooling

  1. Taille des cellules F
  2. Le pas S

BatchNormalization

Importation

import tensorflow as tf
from tensorflow import keras
import matplotlib.pyplot as plt

Hyperparamètres

WIDTH = 150
HEIGHT = 150
NUM_CHANNELS = 3
VALIDATION_SPLIT = 0.2
BATCH_SIZE = 32
LEARNING_RATE = 1e-4
EPOCHS = 10

Télécharger les données

!wget

Préparer les données

generator = keras.preprocessing.image.ImageDataGenerator(
    rescale=1./255,
    validation_split=VALIDATION_SPLIT
)
train_set = generator.flow_from_directory(
    DATA_DIR,
    target_size=(WIDTH, HEIGHT),
    batch_size=BATCH_SIZE,
    class_mode='categorical',
    subset='training',
    shuffle=True
)
val_set = generator.flow_from_directory(
    DATA_DIR,
    target_size=(WIDTH, HEIGHT),
    batch_size=BATCH_SIZE,
    class_mode='categorical',
    subset='validation'
)
num_classes = len(train_set.class_indices)
print("Total classes:", num_classes)

Found 2939 images belonging to 5 classes.
Found 731 images belonging to 5 classes.
Total classes: 5

Conception du modèle

model = keras.Sequential([
    keras.Input(shape=(WIDTH, HEIGHT, NUM_CHANNELS)),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.MaxPooling2D(2, 2),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.MaxPooling2D(2, 2),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.Conv2D(64, (5,5), activation=tf.nn.relu),
    keras.layers.MaxPooling2D(2, 2),
    keras.layers.Flatten(),
    keras.layers.Dense(128, activation=tf.nn.relu),
    keras.layers.Dense(num_classes, activation=tf.nn.softmax)
])
model.summary()

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d (Conv2D)              (None, 146, 146, 64)      4864      
_________________________________________________________________
conv2d_1 (Conv2D)            (None, 142, 142, 64)      102464    
_________________________________________________________________
max_pooling2d (MaxPooling2D) (None, 71, 71, 64)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 67, 67, 64)        102464    
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 63, 63, 64)        102464    
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 31, 31, 64)        0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 27, 27, 64)        102464    
_________________________________________________________________
conv2d_5 (Conv2D)            (None, 23, 23, 64)        102464    
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 11, 11, 64)        0         
_________________________________________________________________
flatten (Flatten)            (None, 7744)              0         
_________________________________________________________________
dense (Dense)                (None, 128)               991360    
_________________________________________________________________
dense_1 (Dense)              (None, 5)                 645       
=================================================================
Total params: 1,509,189
Trainable params: 1,509,189
Non-trainable params: 0
_________________________________________________________________

Paramétrage de l'apprentissage

model.compile(loss="categorical_crossentropy",
              optimizer=keras.optimizers.RMSprop(learning_rate=LEARNING_RATE),
              metrics=["accuracy"])

Entraînement

history = model.fit(train_set, validation_data=val_set, epochs=EPOCHS)

Reporting

Après l'entraînement il est important de visualiser la courbe des métriques 

import matplotlib.pyplot as plt

acc = history.history['accuracy']
loss = history.history['loss']
val_acc = history.history['val_accuracy']
val_loss = history.history['val_loss']
epochs = range(1, len(acc) + 1)

plt.figure(figsize=(8, 4))
plt.plot(epochs, acc, 'bo', label='Training Accuracy')
plt.plot(epochs, val_acc, 'b', label='Validation Accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.title('Training and validation accuracy')
plt.legend(loc='best')
plt.show()

plt.figure(figsize=(8, 4))
plt.plot(epochs, loss, 'bo', label='Training Loss')
plt.plot(epochs, val_loss, 'b', label='Validation Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.title('Training and validation loss')
plt.legend(loc='best')
plt.show()

Ces courbes permettent de savoir si le modèle génralise ou pas.