In [1]:
import tensorflow as tf
from tensorflow import keras
from keras.preprocessing.image import ImageDataGenerator
In [2]:
import warnings
warnings.filterwarnings('ignore')
In [3]:
train_datagen = ImageDataGenerator(rescale = 1./255)
                                  
train_set = train_datagen.flow_from_directory('Data_steel/Train_set/', 
                                               class_mode='categorical',
                                               classes = ['Network','Pearlite_Spheroidite','Martensite', 'Pearlite'],
                                                 target_size = (64, 64),
                                                 batch_size = 32,
                                                 color_mode ='grayscale')
train_set.class_indices
                                    
Found 383 images belonging to 4 classes.
Out[3]:
{'Network': 0, 'Pearlite_Spheroidite': 1, 'Martensite': 2, 'Pearlite': 3}
In [4]:
test_datagen = ImageDataGenerator(rescale = 1./255)
test_set = test_datagen.flow_from_directory('Data_steel/Test_set/',class_mode='categorical',
                                            classes = ['Network','Pearlite_Spheroidite','Martensite', 'Pearlite'],
                                            target_size = (64, 64),
                                            batch_size = 32, color_mode ='grayscale')

                                            
Found 102 images belonging to 4 classes.
In [5]:
cnn = tf.keras.models.Sequential()
In [6]:
cnn.add(tf.keras.layers.Conv2D(filters = 64, kernel_size =3, activation = 'relu', input_shape = [64, 64,1]))
In [7]:
cnn.add(tf.keras.layers.MaxPool2D(pool_size = 2, strides =2))
In [8]:
cnn.add(tf.keras.layers.Conv2D(filters=64, kernel_size=3, activation='relu'))
cnn.add(tf.keras.layers.MaxPool2D(pool_size=2, strides=2))
In [9]:
cnn.add(tf.keras.layers.Flatten())
In [10]:
cnn.add(tf.keras.layers.Dense(units = 128, activation ='relu'))
In [11]:
cnn.add(tf.keras.layers.Dense(units=4, activation='softmax'))
In [12]:
loss_fn = tf.keras.losses.CategoricalCrossentropy(from_logits=True)
optimizer = tf.keras.optimizers.Adam()
In [13]:
cnn.compile(optimizer='adam',
                loss=loss_fn,
                metrics=['accuracy'])
In [14]:
cnn.summary()
Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d (Conv2D)             (None, 62, 62, 64)        640       
                                                                 
 max_pooling2d (MaxPooling2D  (None, 31, 31, 64)       0         
 )                                                               
                                                                 
 conv2d_1 (Conv2D)           (None, 29, 29, 64)        36928     
                                                                 
 max_pooling2d_1 (MaxPooling  (None, 14, 14, 64)       0         
 2D)                                                             
                                                                 
 flatten (Flatten)           (None, 12544)             0         
                                                                 
 dense (Dense)               (None, 128)               1605760   
                                                                 
 dense_1 (Dense)             (None, 4)                 516       
                                                                 
=================================================================
Total params: 1,643,844
Trainable params: 1,643,844
Non-trainable params: 0
_________________________________________________________________
In [15]:
cnn.fit(x = train_set, validation_data = test_set, epochs = 50)
Epoch 1/50
12/12 [==============================] - 8s 566ms/step - loss: 1.3360 - accuracy: 0.4648 - val_loss: 1.2511 - val_accuracy: 0.4216
Epoch 2/50
12/12 [==============================] - 2s 197ms/step - loss: 1.1210 - accuracy: 0.5013 - val_loss: 1.1477 - val_accuracy: 0.4510
Epoch 3/50
12/12 [==============================] - 2s 169ms/step - loss: 1.0632 - accuracy: 0.5379 - val_loss: 1.2203 - val_accuracy: 0.4804
Epoch 4/50
12/12 [==============================] - 2s 150ms/step - loss: 1.0377 - accuracy: 0.5483 - val_loss: 1.1297 - val_accuracy: 0.5000
Epoch 5/50
12/12 [==============================] - 2s 166ms/step - loss: 0.9564 - accuracy: 0.5849 - val_loss: 1.1408 - val_accuracy: 0.4510
Epoch 6/50
12/12 [==============================] - 2s 179ms/step - loss: 0.8752 - accuracy: 0.6501 - val_loss: 1.1001 - val_accuracy: 0.5098
Epoch 7/50
12/12 [==============================] - 1s 110ms/step - loss: 0.8398 - accuracy: 0.6919 - val_loss: 1.0288 - val_accuracy: 0.4804
Epoch 8/50
12/12 [==============================] - 2s 147ms/step - loss: 0.7924 - accuracy: 0.7285 - val_loss: 0.9982 - val_accuracy: 0.5098
Epoch 9/50
12/12 [==============================] - 2s 171ms/step - loss: 0.7071 - accuracy: 0.7467 - val_loss: 1.1161 - val_accuracy: 0.5490
Epoch 10/50
12/12 [==============================] - 2s 144ms/step - loss: 0.5856 - accuracy: 0.8120 - val_loss: 1.2689 - val_accuracy: 0.5392
Epoch 11/50
12/12 [==============================] - 2s 149ms/step - loss: 0.5031 - accuracy: 0.8538 - val_loss: 1.3010 - val_accuracy: 0.5294
Epoch 12/50
12/12 [==============================] - 1s 119ms/step - loss: 0.4331 - accuracy: 0.8773 - val_loss: 1.3814 - val_accuracy: 0.6275
Epoch 13/50
12/12 [==============================] - 2s 142ms/step - loss: 0.3524 - accuracy: 0.9138 - val_loss: 1.1572 - val_accuracy: 0.6176
Epoch 14/50
12/12 [==============================] - 1s 122ms/step - loss: 0.2691 - accuracy: 0.9191 - val_loss: 1.1806 - val_accuracy: 0.6176
Epoch 15/50
12/12 [==============================] - 1s 107ms/step - loss: 0.2196 - accuracy: 0.9556 - val_loss: 1.1092 - val_accuracy: 0.6373
Epoch 16/50
12/12 [==============================] - 1s 102ms/step - loss: 0.1553 - accuracy: 0.9739 - val_loss: 1.4785 - val_accuracy: 0.5490
Epoch 17/50
12/12 [==============================] - 1s 121ms/step - loss: 0.1246 - accuracy: 0.9739 - val_loss: 1.2914 - val_accuracy: 0.6078
Epoch 18/50
12/12 [==============================] - 1s 108ms/step - loss: 0.0909 - accuracy: 0.9817 - val_loss: 1.3437 - val_accuracy: 0.6176
Epoch 19/50
12/12 [==============================] - 2s 131ms/step - loss: 0.0668 - accuracy: 0.9948 - val_loss: 1.2970 - val_accuracy: 0.6176
Epoch 20/50
12/12 [==============================] - 1s 114ms/step - loss: 0.0542 - accuracy: 0.9948 - val_loss: 1.3030 - val_accuracy: 0.6373
Epoch 21/50
12/12 [==============================] - 1s 108ms/step - loss: 0.0471 - accuracy: 0.9948 - val_loss: 1.3960 - val_accuracy: 0.6569
Epoch 22/50
12/12 [==============================] - 2s 152ms/step - loss: 0.0485 - accuracy: 0.9948 - val_loss: 1.4118 - val_accuracy: 0.6569
Epoch 23/50
12/12 [==============================] - 2s 136ms/step - loss: 0.0331 - accuracy: 0.9974 - val_loss: 1.8768 - val_accuracy: 0.6569
Epoch 24/50
12/12 [==============================] - 1s 112ms/step - loss: 0.0297 - accuracy: 0.9974 - val_loss: 1.4452 - val_accuracy: 0.6667
Epoch 25/50
12/12 [==============================] - 2s 135ms/step - loss: 0.0212 - accuracy: 0.9974 - val_loss: 1.2947 - val_accuracy: 0.6373
Epoch 26/50
12/12 [==============================] - 2s 123ms/step - loss: 0.0171 - accuracy: 0.9974 - val_loss: 1.7843 - val_accuracy: 0.6569
Epoch 27/50
12/12 [==============================] - 1s 109ms/step - loss: 0.0111 - accuracy: 0.9974 - val_loss: 1.5953 - val_accuracy: 0.6471
Epoch 28/50
12/12 [==============================] - 2s 133ms/step - loss: 0.0092 - accuracy: 1.0000 - val_loss: 1.8058 - val_accuracy: 0.6176
Epoch 29/50
12/12 [==============================] - 2s 128ms/step - loss: 0.0098 - accuracy: 0.9974 - val_loss: 1.7211 - val_accuracy: 0.6471
Epoch 30/50
12/12 [==============================] - 1s 112ms/step - loss: 0.0076 - accuracy: 1.0000 - val_loss: 1.7474 - val_accuracy: 0.6471
Epoch 31/50
12/12 [==============================] - 2s 124ms/step - loss: 0.0062 - accuracy: 1.0000 - val_loss: 1.6980 - val_accuracy: 0.6569
Epoch 32/50
12/12 [==============================] - 2s 123ms/step - loss: 0.0060 - accuracy: 1.0000 - val_loss: 1.8591 - val_accuracy: 0.6373
Epoch 33/50
12/12 [==============================] - 2s 169ms/step - loss: 0.0047 - accuracy: 1.0000 - val_loss: 1.7689 - val_accuracy: 0.6373
Epoch 34/50
12/12 [==============================] - 1s 118ms/step - loss: 0.0044 - accuracy: 1.0000 - val_loss: 1.8435 - val_accuracy: 0.6471
Epoch 35/50
12/12 [==============================] - 2s 170ms/step - loss: 0.0041 - accuracy: 1.0000 - val_loss: 1.7996 - val_accuracy: 0.6373
Epoch 36/50
12/12 [==============================] - 2s 129ms/step - loss: 0.0036 - accuracy: 1.0000 - val_loss: 1.8851 - val_accuracy: 0.6373
Epoch 37/50
12/12 [==============================] - 1s 113ms/step - loss: 0.0033 - accuracy: 1.0000 - val_loss: 1.8478 - val_accuracy: 0.6471
Epoch 38/50
12/12 [==============================] - 1s 121ms/step - loss: 0.0030 - accuracy: 1.0000 - val_loss: 1.9048 - val_accuracy: 0.6373
Epoch 39/50
12/12 [==============================] - 1s 117ms/step - loss: 0.0029 - accuracy: 1.0000 - val_loss: 1.9382 - val_accuracy: 0.6569
Epoch 40/50
12/12 [==============================] - 2s 156ms/step - loss: 0.0026 - accuracy: 1.0000 - val_loss: 1.8520 - val_accuracy: 0.6471
Epoch 41/50
12/12 [==============================] - 2s 126ms/step - loss: 0.0025 - accuracy: 1.0000 - val_loss: 2.0066 - val_accuracy: 0.6373
Epoch 42/50
12/12 [==============================] - 1s 103ms/step - loss: 0.0023 - accuracy: 1.0000 - val_loss: 1.9022 - val_accuracy: 0.6471
Epoch 43/50
12/12 [==============================] - 1s 103ms/step - loss: 0.0023 - accuracy: 1.0000 - val_loss: 1.9572 - val_accuracy: 0.6373
Epoch 44/50
12/12 [==============================] - 1s 107ms/step - loss: 0.0021 - accuracy: 1.0000 - val_loss: 1.9705 - val_accuracy: 0.6471
Epoch 45/50
12/12 [==============================] - 2s 143ms/step - loss: 0.0020 - accuracy: 1.0000 - val_loss: 1.9667 - val_accuracy: 0.6373
Epoch 46/50
12/12 [==============================] - 1s 109ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 2.0215 - val_accuracy: 0.6471
Epoch 47/50
12/12 [==============================] - 2s 131ms/step - loss: 0.0018 - accuracy: 1.0000 - val_loss: 1.9998 - val_accuracy: 0.6471
Epoch 48/50
12/12 [==============================] - 1s 105ms/step - loss: 0.0016 - accuracy: 1.0000 - val_loss: 2.0160 - val_accuracy: 0.6373
Epoch 49/50
12/12 [==============================] - 1s 111ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 2.0330 - val_accuracy: 0.6471
Epoch 50/50
12/12 [==============================] - 2s 162ms/step - loss: 0.0015 - accuracy: 1.0000 - val_loss: 2.0100 - val_accuracy: 0.6471
Out[15]:
<keras.callbacks.History at 0x26d50273130>
In [16]:
import numpy as np
from keras.preprocessing import image
from keras_preprocessing.image import load_img
from keras_preprocessing.image import img_to_array
#from tensorflow.keras.utils import load_img
#from tensorflow.keras.utils import img_to_array
test_image = load_img('Data_steel/single_prediction/net.tif', target_size = (64, 64),color_mode = 'grayscale')
test_image = img_to_array(test_image)
test_image = np.expand_dims(test_image, axis = 0)                  
result = (cnn.predict(test_image) > 0.3).astype('int')
print(result)
train_set.class_indices

# >0.3 is the decision threshold of prediction
1/1 [==============================] - 0s 184ms/step
[[1 0 0 0]]
Out[16]:
{'Network': 0, 'Pearlite_Spheroidite': 1, 'Martensite': 2, 'Pearlite': 3}