Keras利用卷積神經網絡瑪麗蓮夢露與愛因斯坦的識別Part

目的
突發奇想想會認爲下面這張圖片究竟是瑪麗蓮夢露還是愛因斯坦，主要目的順便實踐練習《Python深度學習》書中的例子，只採用了很小批量的數據，也沒有深究如何提高正確率，解決過擬合的問題。詳細可以參見《python深度學習》第五章前兩節。

數據準備

從百度圖片中找到了風格各異的愛因斯坦的圖片，直接採用下載整個網頁的方式獲取圖片。選的量不多，100張作爲訓練，25張用於驗證。本來是留有測試的數據，不小心刪掉了就跳過在新數據上測試的步驟。（數據量太小也是一個嚴重的問題）
手動刪掉一些不合適的圖片，分別放到train和validation文件夾下的E，M兩個文件中。

構建網絡
建立序列模型，採用這個網絡是因爲之前在一個SAR圖像的識別中表現優異，預測準確率達到96%以上（儘管並不能說明它在區分愛因斯坦和瑪麗蓮夢露也能表現得很好）

import os
from keras import layers
from keras import models
model = models.Sequential()
model.add(layers.Conv2D(64, (3, 3), activation='relu',
      input_shape=(88, 88,3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.summary()
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_1 (Conv2D)            (None, 86, 86, 64)        1792      
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 43, 43, 64)        0         
_________________________________________________________________
conv2d_2 (Conv2D)            (None, 41, 41, 64)        36928     
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 20, 20, 64)        0         
_________________________________________________________________
conv2d_3 (Conv2D)            (None, 18, 18, 128)       73856     
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 9, 9, 128)         0         
_________________________________________________________________
conv2d_4 (Conv2D)            (None, 7, 7, 128)         147584    
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 3, 3, 128)         0         
_________________________________________________________________
flatten_1 (Flatten)          (None, 1152)              0         
_________________________________________________________________
dropout_1 (Dropout)          (None, 1152)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 512)               590336    
_________________________________________________________________
dropout_2 (Dropout)          (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 512)               262656    
_________________________________________________________________
dense_3 (Dense)              (None, 1)                 513       
=================================================================
Total params: 1,113,665
Trainable params: 1,113,665
Non-trainable params: 0

讀入圖片並訓練

base_dir = r'dir\Einstein'
train_dir = os.path.join(base_dir, 'train')
validation_dir = os.path.join(base_dir, 'v')

from keras import optimizers
model.compile(loss='binary_crossentropy',
   optimizer=optimizers.RMSprop(lr=1e-4),
   metrics=['acc'])

from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale=1. / 255)
test_datagen = ImageDataGenerator(rescale=1. / 255)

train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(88,88),
    batch_size=20,
    class_mode='binary')
    
validation_generator = test_datagen.flow_from_directory(
    validation_dir,
    target_size=(88, 88),
    batch_size=20,
    class_mode='binary')
history = model.fit_generator(train_generator,steps_per_epoch=128,epochs=20,
                              validation_data=validation_generator,validation_steps=50)


#保存模型
model.save('EM.h5')

訓練結果
繪製性能曲線

import matplotlib.pyplot as plt

acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(acc) + 1)
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()

l最終結果：
oss: 0.0050 - acc: 0.9996 -
val_loss: 2.4635 - val_acc: 0.6614
結果顯然過擬合了，預測的正確率只有66%。但不妨礙用於用的預測。
預測

from keras.preprocessing import image
import matplotlib.image as mpimg
from keras import models
import numpy as np
img = image.load_img(r'dir\Einstein\EM.jpg',target_size=(88,88,3))
img = np.array(img)
img = img/255
model = models.load_model(r'dir\Einstein\EM.h5')
img = img.reshape(1,88,88,3)
pre = model.predict(img)
print('預測結果：'，pre)


預測結果： [[0.00376787]]

Keras添加的標籤是E（愛因斯坦）文件夾中的爲0，M（瑪麗蓮夢露）爲1。通過網絡最後的sigmoid單元，輸出值爲0.00376787，這個神經網絡十分傾向於認爲這張圖片是愛因斯坦。

嘗試了很多種不同的結構（數據量小訓練也很快），驗證集的正確率一直在70%左右，僅有一次認爲該圖片是瑪麗蓮夢露，其餘結果都認爲這張圖片是愛因斯坦。

結論
在搭建的這樣的簡單的網絡下，更傾向於認爲這種圖片裏的人是愛因斯坦。

不足之處

• 樣本太少
• 過擬合，驗證集的識別正確率不高

參考資料
《python深度學習》