如何提高图像分类器的准确率？答案

【问题标题】：How to increase the accurancy of an image classifier?如何提高图像分类器的准确率？
【发布时间】：2020-03-27 19:34:22
【问题描述】：

我制作了一个模型，可以使用图像数据集（大约 10500 张图像）对 82 个数字进行分类
数据集位于两个文件夹中：
第一个文件夹 train 文件夹在 82 个文件夹中有 8000 个图像
第二个文件夹 test 文件夹在 82 个文件夹中有 2000 个图像
在转到主数据集文件夹之前，我已经在其他 2 个文件夹上测试了模型，效果很好
但在这里我不知道为什么 acc 不会变得更好
请注意，并非我的数据集中的所有文件夹都具有相同数量的图像，图像的分辨率也不相同，但都在 210x50 左右
还请注意，在我第一次尝试使用模型在两个文件夹上对其进行测试时，我制作了两个类别的小数据集，文件夹中的图像数量相同（验证文件夹相同）
下面是我用来创建模型的代码：

from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense
from keras import backend as K


# dimensions of our images.
img_width, img_height = 251, 54
#img_width, img_height = 150, 33

train_data_dir = 'C:/Users/ADEM/Desktop/msi_youssef/PFE/test/numbers/data/train'
validation_data_dir = 'C:/Users/ADEM/Desktop/msi_youssef/PFE/test/numbers/data/valid'
nb_train_samples = 10435
nb_validation_samples = 2051
epochs = 20 # how much time you want to train your model on the data
batch_size = 16

if K.image_data_format() == 'channels_first':
    input_shape = (3, img_width, img_height)
else:
    input_shape = (img_width, img_height, 3)

model = Sequential()
model.add(Conv2D(32, (3, 3), input_shape=input_shape))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Flatten())
model.add(Dense(64))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(1))
model.add(Activation('sigmoid'))

model.compile(loss='binary_crossentropy',optimizer='rmsprop',metrics=['accuracy'])

# this is the augmentation configuration we will use for training
train_datagen = ImageDataGenerator(
    rescale=1. / 255,
    shear_range=0.1,
    zoom_range=0.05,
    horizontal_flip=False)

# this is the augmentation configuration we will use for testing:
# only rescaling
test_datagen = ImageDataGenerator(rescale=1. / 255)

train_generator = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
    validation_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='binary')

model.fit_generator(
    train_generator,
    steps_per_epoch=nb_train_samples // batch_size,
    epochs=epochs,
    validation_data=validation_generator,
    validation_steps=nb_validation_samples // batch_size)

model.save('first_try.h5')

结果如下：

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:74: The name tf.get_default_graph is deprecated. Please use tf.compat.v1.get_default_graph instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:517: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:4138: The name tf.random_uniform is deprecated. Please use tf.random.uniform instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:3976: The name tf.nn.max_pool is deprecated. Please use tf.nn.max_pool2d instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:133: The name tf.placeholder_with_default is deprecated. Please use tf.compat.v1.placeholder_with_default instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.
Instructions for updating:
Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.
WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\optimizers.py:790: The name tf.train.Optimizer is deprecated. Please use tf.compat.v1.train.Optimizer instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:3376: The name tf.log is deprecated. Please use tf.math.log instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\tensorflow_core\python\ops\nn_impl.py:183: where (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.where in 2.0, which has the same broadcast rule as np.where
Found 10435 images belonging to 82 classes.
Found 2051 images belonging to 82 classes.
WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:986: The name tf.assign_add is deprecated. Please use tf.compat.v1.assign_add instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:973: The name tf.assign is deprecated. Please use tf.compat.v1.assign instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:2741: The name tf.Session is deprecated. Please use tf.compat.v1.Session instead.

Epoch 1/20
WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:174: The name tf.get_default_session is deprecated. Please use tf.compat.v1.get_default_session instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:181: The name tf.ConfigProto is deprecated. Please use tf.compat.v1.ConfigProto instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:190: The name tf.global_variables is deprecated. Please use tf.compat.v1.global_variables instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:199: The name tf.is_variable_initialized is deprecated. Please use tf.compat.v1.is_variable_initialized instead.

WARNING:tensorflow:From C:\Users\ADEM\Anaconda3\lib\site-packages\keras\backend\tensorflow_backend.py:206: The name tf.variables_initializer is deprecated. Please use tf.compat.v1.variables_initializer instead.

652/652 [==============================] - 43s 65ms/step - loss: -625.7214 - acc: 0.0143 - val_loss: -632.8458 - val_acc: 0.0112
Epoch 2/20
652/652 [==============================] - 47s 72ms/step - loss: -627.1426 - acc: 0.0143 - val_loss: -632.6816 - val_acc: 0.0113
Epoch 3/20
652/652 [==============================] - 42s 65ms/step - loss: -627.8743 - acc: 0.0143 - val_loss: -633.1438 - val_acc: 0.0113
Epoch 4/20
652/652 [==============================] - 45s 69ms/step - loss: -627.0466 - acc: 0.0142 - val_loss: -632.6816 - val_acc: 0.0108
Epoch 5/20
652/652 [==============================] - 47s 73ms/step - loss: -628.4401 - acc: 0.0143 - val_loss: -632.7599 - val_acc: 0.0118
Epoch 6/20
652/652 [==============================] - 45s 69ms/step - loss: -626.8264 - acc: 0.0143 - val_loss: -631.9844 - val_acc: 0.0108
Epoch 7/20
652/652 [==============================] - 55s 85ms/step - loss: -627.8007 - acc: 0.0141 - val_loss: -636.2931 - val_acc: 0.0103
Epoch 8/20
652/652 [==============================] - 46s 71ms/step - loss: -626.7282 - acc: 0.0144 - val_loss: -633.0968 - val_acc: 0.0123
Epoch 9/20
652/652 [==============================] - 47s 72ms/step - loss: -628.2569 - acc: 0.0143 - val_loss: -633.8959 - val_acc: 0.0113
Epoch 10/20
652/652 [==============================] - 46s 71ms/step - loss: -627.1006 - acc: 0.0144 - val_loss: -629.7360 - val_acc: 0.0113
Epoch 11/20
652/652 [==============================] - 54s 83ms/step - loss: -627.1028 - acc: 0.0142 - val_loss: -636.8650 - val_acc: 0.0098
Epoch 12/20
652/652 [==============================] - 45s 70ms/step - loss: -627.8524 - acc: 0.0143 - val_loss: -627.5894 - val_acc: 0.0118
Epoch 13/20
652/652 [==============================] - 46s 70ms/step - loss: -627.1357 - acc: 0.0142 - val_loss: -631.9687 - val_acc: 0.0118
Epoch 14/20
652/652 [==============================] - 48s 73ms/step - loss: -627.5105 - acc: 0.0146 - val_loss: -638.9724 - val_acc: 0.0118
Epoch 15/20
652/652 [==============================] - 46s 70ms/step - loss: -629.0591 - acc: 0.0136 - val_loss: -630.7622 - val_acc: 0.0103
Epoch 16/20
652/652 [==============================] - 46s 71ms/step - loss: -625.9115 - acc: 0.0147 - val_loss: -630.3392 - val_acc: 0.0098
Epoch 17/20
652/652 [==============================] - 45s 70ms/step - loss: -627.0184 - acc: 0.0144 - val_loss: -636.2304 - val_acc: 0.0123
Epoch 18/20
652/652 [==============================] - 47s 72ms/step - loss: -626.8828 - acc: 0.0144 - val_loss: -634.5618 - val_acc: 0.0118
Epoch 19/20
652/652 [==============================] - 45s 70ms/step - loss: -627.3642 - acc: 0.0140 - val_loss: -629.8300 - val_acc: 0.0118
Epoch 20/20
652/652 [==============================] - 46s 71ms/step - loss: -627.4297 - acc: 0.0142 - val_loss: -637.6797 - val_acc: 0.0108

【问题讨论】：

这是一个二分类还是多分类问题？你有几节课？
根据问题，应该是多类分类问题。应该有82个班。您是否尝试过将 class_mode 从二进制更改为 82 位数字？你可以在这里看到一个简单的例子：medium.com/@vijayabhaskar96/…
请检查我的答案！

标签： python tensorflow machine-learning keras deep-learning

【解决方案1】：

由于您的模型现在正在处理多类问题，因此需要进行一些更改：

损失应该是categorical_crossentropy而不是binary_crossentropy
最终的激活函数应该是 softmax 而不是 sigmoid
如果有 82 个类，那么最后一层应该有 82 个神经元（Dense(82) 而不是 Dense(1)）。

祝你好运！

【讨论】：

我将对此进行测试并回复您，感谢您的帮助和您的时间，希望您有一个幸福的一天！
成功了！但唯一需要更改的是``` binary_crossentropy``` 你需要将其更改为 > 而不是 > 否则你会得到错误：ValueError：检查目标时出错：预期activation_10 的形状为 (80,)，但数组的形状为 (1,)

【解决方案2】：

感谢@danielcahall，我已经更正了模型，现在它可以工作了，我唯一改变的是：

损失应该是sparse_categorical_crossentropy而不是 binary_crossentropy
最终的激活函数应该是softmax而不是sigmoid
最后一层应该有 82 个神经元 Dense(82) 而不是 Dense(1) 如果有 82 个班级。
下面是完整的更正代码：


from keras.preprocessing.image import ImageDataGenerator
from keras.models import Sequential
from keras.layers import Conv2D, MaxPooling2D
from keras.layers import Activation, Dropout, Flatten, Dense
from keras import backend as K


# dimensions of our images.
img_width, img_height = 251, 54
#img_width, img_height = 150, 33

train_data_dir = 'C:/Users/ADEM/Desktop/msi_youssef/PFE/test/numbers/data/train'
validation_data_dir = 'C:/Users/ADEM/Desktop/msi_youssef/PFE/test/numbers/data/valid'
nb_train_samples = 8800 #10435
nb_validation_samples = 1763 #2051
epochs = 20 # how much time you want to train your model on the data
batch_size = 32 #16

if K.image_data_format() == 'channels_first':
    input_shape = (3, img_width, img_height)
else:
    input_shape = (img_width, img_height, 3)

model = Sequential()
model.add(Conv2D(32, (3, 3), input_shape=input_shape))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(32, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(64, (3, 3)))
model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Flatten())
model.add(Dense(64))
model.add(Activation('relu'))
model.add(Dropout(0.5))
model.add(Dense(80)) #1
model.add(Activation('softmax')) #sigmoid

model.compile(loss='sparse_categorical_crossentropy',optimizer='rmsprop',metrics=['accuracy'])#categorical_crossentropy #binary_crossentropy

# this is the augmentation configuration we will use for training
train_datagen = ImageDataGenerator(
    rescale=1. / 255,
    shear_range=0.1,
    zoom_range=0.05,
    horizontal_flip=False)

# this is the augmentation configuration we will use for testing:
# only rescaling
test_datagen = ImageDataGenerator(rescale=1. / 255)

train_generator = train_datagen.flow_from_directory(
    train_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='binary')

validation_generator = test_datagen.flow_from_directory(
    validation_data_dir,
    target_size=(img_width, img_height),
    batch_size=batch_size,
    class_mode='binary')

model.fit_generator(
    train_generator,
    steps_per_epoch=nb_train_samples // batch_size,
    epochs=epochs,
    validation_data=validation_generator,
    validation_steps=nb_validation_samples // batch_size)

model.save('testX_1.h5') #first_try

请注意：我已达到acc: 0.6675，如果您想要更多，您需要增加epochs！

【讨论】：

【解决方案3】：

快速更新用 30 测试了 epochs，现在 acc 是：
acc: 0.7562 - val_loss: 0.1268 - val_acc: 0.9688

【讨论】：