回答編集履歴
1
コード追加
test
CHANGED
|
@@ -11,3 +11,301 @@
|
|
|
11
11
|
Y1_label.append(index1)
|
|
12
12
|
|
|
13
13
|
```
|
|
14
|
+
|
|
15
|
+
【追記】
|
|
16
|
+
|
|
17
|
+
上記の修正の他に、明らかな書き間違いと思われるところも全部直した下記のコードで、適当な2種類に画像を分類するデータセットを使って、Google Colabで実行したところ、正常に学習され、「print」文で表示される各数値もそれなりの正しそうな値が表示されました
|
|
18
|
+
|
|
19
|
+
```python
|
|
20
|
+
|
|
21
|
+
#import keras
|
|
22
|
+
|
|
23
|
+
import glob
|
|
24
|
+
|
|
25
|
+
import numpy as np
|
|
26
|
+
|
|
27
|
+
from sklearn.model_selection import train_test_split
|
|
28
|
+
|
|
29
|
+
from tensorflow.keras.preprocessing.image import load_img, img_to_array
|
|
30
|
+
|
|
31
|
+
from tensorflow.keras.models import Sequential
|
|
32
|
+
|
|
33
|
+
from tensorflow.keras.layers import Conv2D, MaxPooling2D
|
|
34
|
+
|
|
35
|
+
from tensorflow.keras.layers import Dense, Dropout, Flatten
|
|
36
|
+
|
|
37
|
+
from tensorflow.keras.utils import to_categorical
|
|
38
|
+
|
|
39
|
+
from tensorflow.keras.optimizers import Adam
|
|
40
|
+
|
|
41
|
+
import matplotlib.pyplot as plt
|
|
42
|
+
|
|
43
|
+
import time
|
|
44
|
+
|
|
45
|
+
import pickle
|
|
46
|
+
|
|
47
|
+
from sklearn.metrics import confusion_matrix
|
|
48
|
+
|
|
49
|
+
from sklearn import metrics
|
|
50
|
+
|
|
51
|
+
from sklearn.metrics import precision_score
|
|
52
|
+
|
|
53
|
+
from sklearn.metrics import recall_score
|
|
54
|
+
|
|
55
|
+
|
|
56
|
+
|
|
57
|
+
'''
|
|
58
|
+
|
|
59
|
+
def set_random_seed(seed):
|
|
60
|
+
|
|
61
|
+
random.seed(seed)
|
|
62
|
+
|
|
63
|
+
np.random.seed (seed)
|
|
64
|
+
|
|
65
|
+
tf.set_random_seed(seed)
|
|
66
|
+
|
|
67
|
+
'''
|
|
68
|
+
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
train_data_path = 'dataset'
|
|
72
|
+
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
test_data_path = 'dataset1'
|
|
76
|
+
|
|
77
|
+
|
|
78
|
+
|
|
79
|
+
image_size = 28
|
|
80
|
+
|
|
81
|
+
|
|
82
|
+
|
|
83
|
+
color_setting = 3
|
|
84
|
+
|
|
85
|
+
|
|
86
|
+
|
|
87
|
+
folder = ['normal', 'abnormal']
|
|
88
|
+
|
|
89
|
+
folders = ['normal', 'abnormal']
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+
|
|
93
|
+
class_number = len(folder)
|
|
94
|
+
|
|
95
|
+
print('今回のデータで分類するクラス数は「', str(class_number), '」です。')
|
|
96
|
+
|
|
97
|
+
|
|
98
|
+
|
|
99
|
+
X_image = []
|
|
100
|
+
|
|
101
|
+
Y_label = []
|
|
102
|
+
|
|
103
|
+
X1_image = []
|
|
104
|
+
|
|
105
|
+
Y1_label = []
|
|
106
|
+
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
for index, name in enumerate(folder):
|
|
110
|
+
|
|
111
|
+
read_data = train_data_path + '/' + name
|
|
112
|
+
|
|
113
|
+
files = glob.glob(read_data + '/*.jpg')
|
|
114
|
+
|
|
115
|
+
print('--- 読み込んだデータセットは', read_data, 'です。')
|
|
116
|
+
|
|
117
|
+
for i, file in enumerate(files):
|
|
118
|
+
|
|
119
|
+
if color_setting == 1:
|
|
120
|
+
|
|
121
|
+
img = load_img(file, color_mode = 'grayscale' ,target_size=(image_size, image_size))
|
|
122
|
+
|
|
123
|
+
elif color_setting == 3:
|
|
124
|
+
|
|
125
|
+
img = load_img(file, color_mode = 'rgb' ,target_size=(image_size, image_size))
|
|
126
|
+
|
|
127
|
+
array = img_to_array(img)
|
|
128
|
+
|
|
129
|
+
X_image.append(array)
|
|
130
|
+
|
|
131
|
+
Y_label.append(index)
|
|
132
|
+
|
|
133
|
+
|
|
134
|
+
|
|
135
|
+
for index1, name in enumerate(folders):
|
|
136
|
+
|
|
137
|
+
read_data1 = test_data_path + '/' + name
|
|
138
|
+
|
|
139
|
+
files1 = glob.glob(read_data1 + '/*.jpg')
|
|
140
|
+
|
|
141
|
+
print('--- 読み込んだデータセットは', read_data1, 'です。')
|
|
142
|
+
|
|
143
|
+
for j, file1 in enumerate(files1):
|
|
144
|
+
|
|
145
|
+
if color_setting == 1:
|
|
146
|
+
|
|
147
|
+
img1 = load_img(file1, color_mode = 'grayscale' ,target_size=(image_size, image_size))
|
|
148
|
+
|
|
149
|
+
elif color_setting == 3:
|
|
150
|
+
|
|
151
|
+
img1 = load_img(file1, color_mode = 'rgb' ,target_size=(image_size, image_size))
|
|
152
|
+
|
|
153
|
+
array = img_to_array(img1)
|
|
154
|
+
|
|
155
|
+
X1_image.append(array)
|
|
156
|
+
|
|
157
|
+
#Y1_label.append(index)
|
|
158
|
+
|
|
159
|
+
Y1_label.append(index1)
|
|
160
|
+
|
|
161
|
+
|
|
162
|
+
|
|
163
|
+
X_image = np.array(X_image)
|
|
164
|
+
|
|
165
|
+
Y_label = np.array(Y_label)
|
|
166
|
+
|
|
167
|
+
|
|
168
|
+
|
|
169
|
+
X1_image = np.array(X1_image)
|
|
170
|
+
|
|
171
|
+
Y1_label = np.array(Y1_label)
|
|
172
|
+
|
|
173
|
+
|
|
174
|
+
|
|
175
|
+
X_image = X_image.astype('float32') / 255
|
|
176
|
+
|
|
177
|
+
Y_label = to_categorical(Y_label, class_number)
|
|
178
|
+
|
|
179
|
+
|
|
180
|
+
|
|
181
|
+
X1_image = X1_image.astype('float32') / 255
|
|
182
|
+
|
|
183
|
+
Y1_label = to_categorical(Y1_label, class_number)
|
|
184
|
+
|
|
185
|
+
|
|
186
|
+
|
|
187
|
+
x_train = X_image
|
|
188
|
+
|
|
189
|
+
y_train = Y_label
|
|
190
|
+
|
|
191
|
+
x_test = X1_image
|
|
192
|
+
|
|
193
|
+
y_test = Y1_label
|
|
194
|
+
|
|
195
|
+
|
|
196
|
+
|
|
197
|
+
#4 機械学習（人工知能）モデルの作成 – 畳み込みニューラルネットワーク（CNN）・学習の実行等
|
|
198
|
+
|
|
199
|
+
|
|
200
|
+
|
|
201
|
+
model = Sequential()
|
|
202
|
+
|
|
203
|
+
model.add(Conv2D(16, (3, 3), padding='same',
|
|
204
|
+
|
|
205
|
+
input_shape=(image_size, image_size, color_setting), activation='relu'))
|
|
206
|
+
|
|
207
|
+
model.add(MaxPooling2D(pool_size=(2, 2)))
|
|
208
|
+
|
|
209
|
+
model.add(Conv2D(128, (3, 3), padding='same', activation='relu'))
|
|
210
|
+
|
|
211
|
+
model.add(Conv2D(256, (3, 3), padding='same', activation='relu'))
|
|
212
|
+
|
|
213
|
+
model.add(MaxPooling2D(pool_size=(2, 2)))
|
|
214
|
+
|
|
215
|
+
model.add(Dropout(0.5))
|
|
216
|
+
|
|
217
|
+
model.add(Flatten())
|
|
218
|
+
|
|
219
|
+
model.add(Dense(128, activation='relu'))
|
|
220
|
+
|
|
221
|
+
model.add(Dropout(0.25))
|
|
222
|
+
|
|
223
|
+
model.add(Dense(class_number, activation='softmax'))
|
|
224
|
+
|
|
225
|
+
|
|
226
|
+
|
|
227
|
+
model.summary()
|
|
228
|
+
|
|
229
|
+
|
|
230
|
+
|
|
231
|
+
model.compile(loss='categorical_crossentropy',
|
|
232
|
+
|
|
233
|
+
optimizer=Adam(),
|
|
234
|
+
|
|
235
|
+
metrics=['accuracy'])
|
|
236
|
+
|
|
237
|
+
|
|
238
|
+
|
|
239
|
+
start_time = time.time()
|
|
240
|
+
|
|
241
|
+
|
|
242
|
+
|
|
243
|
+
history = model.fit(x_train,y_train, batch_size=128, epochs=5, verbose=1, validation_data=(x_test, y_test))
|
|
244
|
+
|
|
245
|
+
|
|
246
|
+
|
|
247
|
+
print (metrics.confusion_matrix(y_test.argmax(axis=1), model.predict(x_test).argmax(axis=1)))
|
|
248
|
+
|
|
249
|
+
|
|
250
|
+
|
|
251
|
+
print (recall_score(y_test.argmax(axis=1), model.predict(x_test).argmax(axis=1)))
|
|
252
|
+
|
|
253
|
+
|
|
254
|
+
|
|
255
|
+
print (precision_score(y_test.argmax(axis=1), model.predict(x_test).argmax(axis=1)))
|
|
256
|
+
|
|
257
|
+
|
|
258
|
+
|
|
259
|
+
plt.plot(history.history['accuracy'])
|
|
260
|
+
|
|
261
|
+
plt.plot(history.history['val_accuracy'])
|
|
262
|
+
|
|
263
|
+
plt.title('Model accuracy')
|
|
264
|
+
|
|
265
|
+
plt.ylabel('Accuracy')
|
|
266
|
+
|
|
267
|
+
plt.xlabel('Epoch')
|
|
268
|
+
|
|
269
|
+
plt.grid()
|
|
270
|
+
|
|
271
|
+
plt.legend(['Train', 'Validation'], loc='upper left')
|
|
272
|
+
|
|
273
|
+
plt.show()
|
|
274
|
+
|
|
275
|
+
|
|
276
|
+
|
|
277
|
+
plt.plot(history.history['loss'])
|
|
278
|
+
|
|
279
|
+
plt.plot(history.history['val_loss'])
|
|
280
|
+
|
|
281
|
+
plt.title('Model loss')
|
|
282
|
+
|
|
283
|
+
plt.ylabel('Loss')
|
|
284
|
+
|
|
285
|
+
plt.xlabel('Epoch')
|
|
286
|
+
|
|
287
|
+
plt.grid()
|
|
288
|
+
|
|
289
|
+
plt.legend(['Train', 'Validation'], loc='upper left')
|
|
290
|
+
|
|
291
|
+
plt.show()
|
|
292
|
+
|
|
293
|
+
|
|
294
|
+
|
|
295
|
+
open('cnn_model.json','w').write(model.to_json())
|
|
296
|
+
|
|
297
|
+
model.save_weights('cnn_weights.h5')
|
|
298
|
+
|
|
299
|
+
#model.save('cnn_model_weight.h5') #モデル構造と重みを1つにまとめることもできます
|
|
300
|
+
|
|
301
|
+
|
|
302
|
+
|
|
303
|
+
score = model.evaluate(x_test, y_test, verbose=0)
|
|
304
|
+
|
|
305
|
+
print('Loss:', score[0], '（損失関数値 - 0に近いほど正解に近い）')
|
|
306
|
+
|
|
307
|
+
print('Accuracy:', score[1] * 100, '%', '（精度 - 100% に近いほど正解に近い）')
|
|
308
|
+
|
|
309
|
+
print('Computation time（計算時間）:{0:.3f} sec（秒）'.format(time.time() - start_time))
|
|
310
|
+
|
|
311
|
+
```
|