컨볼루션 뉴럴 네트워크(CNN)를 위한 Tensorflow Keras 기본 패키지 불러오기

이미지 자료를 대상을 컨볼루션 뉴럴 네트워크를 실행하기 위해서 기본적으로 필요한 패키지는 다음과 같다.

Convolution Neural Network, CNN 구조

1.  CNN 기본 패키지 로드하기

from tensorflow import keras
# from tensorflow.keras.layers import Input, Conv2D, MaxPooling2D, BatchNormalization, Dropout, Flatten, Dense

 

위의 그림과 같은 CNN 모델을 케라스 패키지를 이용해서 정의해보자

크게 보면 입력데이터 레이어, 컨볼루션 레이어 2개 , Flatten 레이어 1개, Dense 레이어 2개, 출력데이터 레이어 등 

5개 부분으로 구성되어 있는 것을 알 수 있다.

 

2. 입력데이터의 크기을 정의한다.

img_height = 128
img_width = 128
img_channel = 3

3. CNN 기본 모델을 정의한다.

input_shape = (img_height, img_width, img_channel)

i = keras.Input(shape=input_shape) 

conv1 = keras.layers.Conv2D(32, kernel_size = (3,3),  activation='relu', padding="same")(i)
pool1 = keras.layers.MaxPooling2D((2,2))(conv1)
norm1 = keras.layers.BatchNormalization(axis = -1)(pool1)
dropout1 = keras.layers.Dropout(rate=0.2)(norm1)

conv2 = keras.layers.Conv2D(32, kernel_size = (3,3),  activation='relu', padding="same")(dropout1 )
pool2 = keras.layers.MaxPooling2D((2,2))(conv2)
norm2 = keras.layers.BatchNormalization(axis = -1)(pool2)
dropout2 = keras.layers.Dropout(rate=0.2)(norm2)

flat = keras.layers.Flatten()(dropout2)

hidden1 = keras.layers.Dense(512, activation='relu')(flat)
norm3 = keras.layers.BatchNormalization(axis = -1)(hidden1)
drop3 = keras.layers.Dropout(rate=0.2)(norm3)

hidden2 = keras.layers.Dense(256, activation='relu')(drop3)
norm4 = keras.layers.BatchNormalization(axis = -1)(hidden2)
drop4 = keras.layers.Dropout(rate=0.2)(norm4)

out = keras.layers.Dense(2, activation='sigmoid')(drop4) 

model = keras.Model(inputs=i, outputs=out)
model.compile(optimizer='adam',
                loss='categorical_crossentropy',   #Check between binary_crossentropy and categorical_crossentropy
                metrics=['accuracy'])
print(model.summary())

모델 구조 결과

Model: "model_5"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 input_18 (InputLayer)       [(None, 64, 64, 3)]       0         
                                                                 
 conv2d_25 (Conv2D)          (None, 64, 64, 32)        896       
                                                                 
 max_pooling2d_24 (MaxPoolin  (None, 32, 32, 32)       0         
 g2D)                                                            
                                                                 
 batch_normalization_29 (Bat  (None, 32, 32, 32)       128       
 chNormalization)                                                
                                                                 
 dropout_29 (Dropout)        (None, 32, 32, 32)        0         
                                                                 
 conv2d_26 (Conv2D)          (None, 32, 32, 32)        9248      
                                                                 
 max_pooling2d_25 (MaxPoolin  (None, 16, 16, 32)       0         
 g2D)                                                            
                                                                 
 batch_normalization_30 (Bat  (None, 16, 16, 32)       128       
 chNormalization)                                                
                                                                 
 dropout_30 (Dropout)        (None, 16, 16, 32)        0         
                                                                 
 flatten_7 (Flatten)         (None, 8192)              0         
                                                                 
 dense_15 (Dense)            (None, 512)               4194816   
                                                                 
 batch_normalization_31 (Bat  (None, 512)              2048      
 chNormalization)                                                
                                                                 
 dropout_31 (Dropout)        (None, 512)               0         
                                                                 
 dense_16 (Dense)            (None, 256)               131328    
                                                                 
 batch_normalization_32 (Bat  (None, 256)              1024      
 chNormalization)                                                
                                                                 
 dropout_32 (Dropout)        (None, 256)               0         
                                                                 
 dense_17 (Dense)            (None, 2)                 514       
                                                                 
=================================================================
Total params: 4,340,130
Trainable params: 4,338,466
Non-trainable params: 1,664
_________________________________________________________________
None