def model (train_data_whole,train_labels_whole,test_data,test_labels,opt,epoch,batch_size_factor,num_classes,result_path,feature_extraction,feature_extractor_parameters):
train_data, valid_data, train_labels, valid_labels = train_test_split(train_data_whole, train_labels_whole,test_size=0.1, random_state=13)
train_labels_one_hot=data_preprocessing.labels_convert_one_hot(train_labels)
valid_labels_one_hot=data_preprocessing.labels_convert_one_hot(valid_labels)
test_labels_one_hot=data_preprocessing.labels_convert_one_hot(test_labels)
batch_size=round(train_data.shape[0]/batch_size_factor)
input_shape= (224,224,3)
vgg_model = VGG16(weights='imagenet',
include_top=False,
input_shape=input_shape)
# Creating dictionary that maps layer names to the layers
layer_dict = dict([(layer.name, layer) for layer in vgg_model.layers])
# Getting output tensor of the last VGG layer that we want to include
x = layer_dict['block2_pool'].output
x = Conv2D(filters=64, kernel_size=(3, 3), activation='relu',padding='same')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Conv2D(filters=128, kernel_size=(3, 3), activation='relu',padding='same')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Conv2D(filters=128, kernel_size=(3, 3), activation='relu',padding='same')(x)
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Flatten()(x)
x = Dense(4096, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(4096, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(2, activation='softmax')(x)
# Creating new model. Please note that this is NOT a Sequential() model.
classification_model = Model(input=vgg_model.input, output=x)
for layer in classification_model.layers[:7]:
layer.trainable = True
es=keras.callbacks.EarlyStopping(monitor='val_acc',
min_delta=0,
patience=5000,
verbose=1, mode='auto')
mc = ModelCheckpoint(os.path.join(result_path,'best_model.h5'), monitor='val_acc', mode='auto', save_best_only=True)
classification_model.compile(loss='mean_squared_error',optimizer=opt,metrics=['accuracy'])
classification_train = classification_model.fit(train_data, train_labels_one_hot, batch_size=batch_size, epochs=epoch,
verbose=1, validation_data=(valid_data, valid_labels_one_hot),callbacks=[mc,es])
best_model=load_model(os.path.join(result_path,'best_model.h5'))
file_name=os.path.split(result_path)[1]
date=os.path.split(os.path.split(result_path)[0])[1]
classification_model.save(os.path.join(result_path,date+'_'+file_name+'_'+'VGGpretrained_model.h5'))
if feature_extraction==1:
feature_extractor_parameters['CNN_model']=classification_model
CNN_feature_extractor.CNN_feature_extraction_classsification(feature_extractor_parameters,result_path)
return
model_evaluation.testing_and_printing(classification_model,classification_train,best_model,test_data, test_labels_one_hot, 'VGG_pretrained', result_path,epoch)
def model(train_data_whole, train_labels_whole, test_data, test_labels, opt,
epoch, batch_size_factor, num_classes, result_path,
feature_extraction, feature_extractor_parameters):
train_data, valid_data, train_labels, valid_labels = train_test_split(
train_data_whole, train_labels_whole, test_size=0.1, random_state=13)
train_labels_one_hot = data_preprocessing.labels_convert_one_hot(
train_labels)
valid_labels_one_hot = data_preprocessing.labels_convert_one_hot(
valid_labels)
test_labels_one_hot = data_preprocessing.labels_convert_one_hot(
test_labels)
'''
train_data=data_preprocessing.depth_reshapeing(train_data)
test_data=data_preprocessing.depth_reshapeing(test_data)
valid_data=data_preprocessing.depth_reshapeing(valid_data)
train_data = data_preprocessing.size_editing(train_data, 224)
valid_data= data_preprocessing.size_editing(valid_data, 224)
test_data = data_preprocessing.size_editing(test_data, 224)
'''
batch_size = round(train_data.shape[0] / batch_size_factor)
input_shape = (224, 224, 3)
densenet121_model = keras.applications.densenet.DenseNet121(
include_top=False,
weights='imagenet',
input_shape=input_shape,
pooling=None,
classes=num_classes)
layer_dict = dict([(layer.name, layer)
for layer in densenet121_model.layers])
# Getting output tensor of the last VGG layer that we want to include
x = layer_dict[list(layer_dict.keys())[-1]].output
x = MaxPooling2D(pool_size=(2, 2))(x)
x = Flatten()(x)
x = Dense(4096, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(4096, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(num_classes, activation='softmax')(x)
classification_model = Model(input=densenet121_model.input, output=x)
for layer in classification_model.layers:
layer.trainable = True
classification_model.compile(loss='mean_squared_error',
optimizer=opt,
metrics=['accuracy'])
es = keras.callbacks.EarlyStopping(monitor='val_acc',
min_delta=0,
patience=500,
verbose=1,
mode='auto')
mc = ModelCheckpoint(os.path.join(result_path, 'best_model.h5'),
monitor='val_acc',
mode='auto',
save_best_only=True)
classification_train = classification_model.fit(
train_data,
train_labels_one_hot,
batch_size=batch_size,
epochs=epoch,
verbose=1,
validation_data=(valid_data, valid_labels_one_hot),
callbacks=[es, mc])
file_name = os.path.split(result_path)[1]
date = os.path.split(os.path.split(result_path)[0])[1]
classification_model.save(
os.path.join(result_path,
date + '_' + file_name + '_' + 'DenseNet121.h5'))
#best_model=load_model(os.path.join(result_path,'best_model.h5'))
best_model = classification_model
if feature_extraction == 1:
feature_extractor_parameters['CNN_model'] = classification_model
CNN_feature_extractor.CNN_feature_extraction_classsification(
feature_extractor_parameters, result_path)
return
model_evaluation.testing_and_printing(classification_model,
classification_train, best_model,
test_data, test_labels_one_hot,
'InceptionResNetV2', result_path,
epoch)
def model(train_data_whole, train_labels_whole, test_data, test_labels, opt,
epoch, batch_size_factor, num_classes, result_path,
feature_extraction, feature_extractor_parameters):
train_data, valid_data, train_labels, valid_labels = train_test_split(
train_data_whole, train_labels_whole, test_size=0.1, random_state=13)
train_labels_one_hot = data_preprocessing.labels_convert_one_hot(
train_labels)
valid_labels_one_hot = data_preprocessing.labels_convert_one_hot(
valid_labels)
test_labels_one_hot = data_preprocessing.labels_convert_one_hot(
test_labels)
batch_size = round(train_data.shape[0] / batch_size_factor)
classification_model = Sequential()
# C1 Convolutional Layer
classification_model.add(
layers.Conv2D(6,
kernel_size=(5, 5),
strides=(1, 1),
activation='relu',
input_shape=(train_data.shape[1], train_data.shape[2],
train_data.shape[3]),
padding='same'))
classification_model.add(Dropout(0.7))
# S2 Pooling Layer
classification_model.add(
layers.AveragePooling2D(pool_size=(2, 2),
strides=(1, 1),
padding='valid'))
# C3 Convolutional Layer
classification_model.add(
layers.Conv2D(16,
kernel_size=(5, 5),
strides=(1, 1),
padding='valid',
activation='relu'))
classification_model.add(Dropout(0.7))
# S4 Pooling Layer
classification_model.add(
layers.AveragePooling2D(pool_size=(2, 2),
strides=(2, 2),
padding='valid'))
# C5 Fully Connected Convolutional Layer
classification_model.add(
layers.Conv2D(120,
kernel_size=(5, 5),
strides=(1, 1),
padding='valid',
activation='relu'))
classification_model.add(Dropout(0.8))
#Flatten the CNN output so that we can connect it with fully connected layers
classification_model.add(layers.Flatten())
# FC6 Fully Connected Layer
classification_model.add(layers.Dense(84, activation='relu'))
classification_model.add(Dropout(0.8))
#Output Layer with softmax activation
classification_model.add(layers.Dense(num_classes, activation='softmax'))
classification_model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=opt,
metrics=['accuracy'])
es = keras.callbacks.EarlyStopping(monitor='val_acc',
min_delta=0,
patience=100,
verbose=1,
mode='auto')
mc = ModelCheckpoint(os.path.join(result_path, 'best_model.h5'),
monitor='val_acc',
mode='auto',
save_best_only=True)
classification_train = classification_model.fit(
train_data,
train_labels_one_hot,
batch_size=batch_size,
epochs=epoch,
verbose=1,
validation_data=(valid_data, valid_labels_one_hot),
callbacks=[es, mc])
best_model = load_model(os.path.join(result_path, 'best_model.h5'))
file_name = os.path.split(result_path)[1]
date = os.path.split(os.path.split(result_path)[0])[1]
classification_model.save(
os.path.join(result_path,
date + '_' + file_name + '_' + 'leNet_model.h5'))
if feature_extraction == 1:
feature_extractor_parameters['CNN_model'] = classification_model
CNN_feature_extractor.CNN_feature_extraction_classsification(
train_data_whole, train_labels_whole, test_data, test_labels,
feature_extractor_parameters, result_path)
return
model_evaluation.testing_and_printing(classification_model,
classification_train, best_model,
test_data, test_labels_one_hot,
'LeNet', result_path, epoch)
def model(train_data_whole, train_labels_whole, test_data, test_labels, opt,
epoch, batch_size_factor, num_classes, result_path,
feature_extraction, feature_extractor_parameters):
train_data, valid_data, train_labels, valid_labels = train_test_split(
train_data_whole, train_labels_whole, test_size=0.1, random_state=13)
train_labels_one_hot = data_preprocessing.labels_convert_one_hot(
train_labels)
valid_labels_one_hot = data_preprocessing.labels_convert_one_hot(
valid_labels)
test_labels_one_hot = data_preprocessing.labels_convert_one_hot(
test_labels)
batch_size = round(train_data.shape[0] / batch_size_factor)
#Instantiate an empty model
classification_model = Sequential()
# 1st Convolutional Layer
classification_model.add(
Conv2D(filters=64,
input_shape=(train_data.shape[1], train_data.shape[2],
train_data.shape[3]),
activation='relu',
kernel_size=(3, 3),
strides=(1, 1),
padding='same'))
classification_model.add(
Conv2D(filters=64,
activation='relu',
kernel_size=(3, 3),
strides=(1, 1),
padding='same'))
classification_model.add(Dropout(0.4))
# Max Pooling
classification_model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
# 2nd Convolutional Layer
classification_model.add(
Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=128,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(Dropout(0.4))
# Max Pooling
classification_model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
# 3rd Convolutional Layer
classification_model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=256,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(Dropout(0.4))
# Max Pooling
classification_model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2)))
# 4th Convolutional Layer
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(Dropout(0.4))
# 5th Convolutional Layer
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(
Conv2D(filters=512,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation='relu'))
classification_model.add(Dropout(0.4))
# Max Pooling
classification_model.add(
MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='valid'))
# Passing it to a Fully Connected layer
classification_model.add(Flatten())
# 1st Fully Connected Layer
classification_model.add(Dense(4096, activation='relu'))
# Add Dropout to prevent overfitting
classification_model.add(Dropout(0.5))
# 2nd Fully Connected Layer
classification_model.add(Dense(4096, activation='relu'))
# Add Dropout
classification_model.add(Dropout(0.5))
# Output Layer
classification_model.add(Dense(num_classes, activation='softmax'))
classification_model.summary()
classification_model.compile(loss='mean_squared_error',
optimizer=opt,
metrics=['accuracy'])
es = keras.callbacks.EarlyStopping(monitor='val_acc',
min_delta=0,
patience=1000,
verbose=1,
mode='auto')
mc = ModelCheckpoint(os.path.join(result_path, 'best_model.h5'),
monitor='val_acc',
mode='auto',
save_best_only=True)
classification_train = classification_model.fit(
train_data,
train_labels_one_hot,
batch_size=batch_size,
epochs=epoch,
verbose=1,
validation_data=(valid_data, valid_labels_one_hot),
callbacks=[mc, es])
best_model = load_model(os.path.join(result_path, 'best_model.h5'))
file_name = os.path.split(result_path)[1]
date = os.path.split(os.path.split(result_path)[0])[1]
classification_model.save(
os.path.join(result_path,
date + '_' + file_name + '_' + 'VGG_model.h5'))
if feature_extraction == 1:
feature_extractor_parameters['CNN_model'] = classification_model
CNN_feature_extractor.CNN_feature_extraction_classsification(
feature_extractor_parameters, result_path)
return
model_evaluation.testing_and_printing(classification_model,
classification_train, best_model,
test_data, test_labels_one_hot,
'VGG', result_path, epoch)
def model(train_data_whole, train_labels_whole, test_data, test_labels, opt,
epoch, batch_size_factor, num_classes, result_path,
feature_extraction, feature_extractor_parameters):
train_data, valid_data, train_labels, valid_labels = train_test_split(
train_data_whole, train_labels_whole, test_size=0.2, random_state=13)
train_labels_one_hot = data_preprocessing.labels_convert_one_hot(
train_labels)
valid_labels_one_hot = data_preprocessing.labels_convert_one_hot(
valid_labels)
test_labels_one_hot = data_preprocessing.labels_convert_one_hot(
test_labels)
batch_size = round(train_data.shape[0] / batch_size_factor)
classification_model = Sequential()
classification_model.add(
Conv2D(32,
kernel_size=(3, 3),
padding='same',
activation='relu',
input_shape=(train_data.shape[1], train_data.shape[2],
train_data.shape[3])))
#classification_model.add(BatchNormalization())
classification_model.add(MaxPooling2D((2, 2), padding='same'))
classification_model.add(Dropout(0.5))
classification_model.add(Conv2D(64, (3, 3), padding='same'))
classification_model.add(LeakyReLU(alpha=0.1))
#classification_model.add(BatchNormalization())
classification_model.add(MaxPooling2D(pool_size=(2, 2), padding='same'))
classification_model.add(Dropout(0.5))
classification_model.add(Conv2D(128, (3, 3), padding='same'))
classification_model.add(LeakyReLU(alpha=0.1))
#classification_model.add(BatchNormalization())
classification_model.add(MaxPooling2D(pool_size=(2, 2), padding='same'))
classification_model.add(Dropout(0.5))
classification_model.add(Flatten())
classification_model.add(Dense(128))
classification_model.add(LeakyReLU(alpha=0.1))
#classification_model.add(BatchNormalization())
classification_model.add(Dense(128, ))
classification_model.add(Dropout(0.5))
classification_model.add(LeakyReLU(alpha=0.1))
classification_model.add(Dense(num_classes, activation='softmax'))
classification_model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=opt,
metrics=['accuracy'])
es = keras.callbacks.EarlyStopping(monitor='val_acc',
min_delta=0,
patience=50,
verbose=1,
mode='auto',
baseline=0.9)
mc = ModelCheckpoint(os.path.join(result_path, 'best_model.h5'),
monitor='val_acc',
mode='auto',
save_best_only=True)
classification_train = classification_model.fit(
train_data,
train_labels_one_hot,
batch_size=batch_size,
epochs=epoch,
verbose=1,
validation_data=(valid_data, valid_labels_one_hot),
callbacks=[es, mc])
best_model = load_model(os.path.join(result_path, 'best_model.h5'))
file_name = os.path.split(result_path)[1]
date = os.path.split(os.path.split(result_path)[0])[1]
classification_model.save(
os.path.join(result_path,
date + '_' + file_name + '_' + 'simple_model.h5'))
if feature_extraction == 1:
feature_extractor_parameters['CNN_model'] = classification_model
CNN_feature_extractor.CNN_feature_extraction_classsification(
train_data_whole, train_labels_whole, test_data, test_labels,
feature_extractor_parameters, result_path)
return
model_evaluation.testing_and_printing(classification_model,
classification_train, best_model,
test_data, test_labels_one_hot,
'simple_architecture', result_path,
epoch)