如何改进数字识别模型？

Question

我正在制作一个简单的应用程序来实现数字识别。问题是，它在mnist数据集上工作得很好，但对从谷歌下载的随机图像的预测却很糟糕。我应该做什么来提高我的模型的准确性？

另一个问题:有人建议在模型中添加更多层。如果是这样，我如何向我的模型添加更多层？

训练模型的文件：

from __future__ import print_function
import keras
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D
from keras import backend as K
import cv2
import matplotlib.pyplot as plt
from keras.models import load_model
import pickle
import h5py
import numpy as np
from keras.callbacks import ModelCheckpoint
batch_size = 128
num_classes = 10
epochs = 12

# input image dimensions
img_rows, img_cols = 28, 28

model=load_model('my_model.h5')



the data, split between train and test sets
(x_train, y_train), (x_test, y_test) = mnist.load_data()
print('initial shape')
print('x_test ',x_test.shape)
print('y_test ',y_test.shape)


# print(x_test);
# print(y_test.shape)

if K.image_data_format() == 'channels_first':
    print('reshape1')
    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)
    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)
else:
    print('reshape2')
    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)
    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255
x_test /= 255
print('x_train shape:', x_train.shape)
print('x_test shape: ' , x_test.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

print('x_test final : ')
print(x_test)

# convert class vectors to binary class matrices
y_train = keras.utils.to_categorical(y_train, num_classes)
y_test = keras.utils.to_categorical(y_test, num_classes)

model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.10))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.20))
model.add(Dense(256,activation='relu'))
model.add(Dropout(0.40))

model.add(Dense(num_classes, activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])

model.fit(x_train, y_train,
          batch_size=batch_size,
          epochs=epochs,
          verbose=1,
          validation_data=(x_test, y_test))

model.save('my_model.h5')
score = model.evaluate(x_test, y_test, verbose=0)


# print(x_test.shape)
# print('\n')
# print(y_test.shape)

print('Test loss:', score[0])
print('Test accuracy:', score[1])

下面的代码用于使用我自己的图像测试我的模型

image = cv2.imread("2.jpg")
img_rows, img_cols = 28, 28

x_test1 = cv2.resize(image, (28, 28)) 

x_test1 = cv2.cvtColor(x_test1,cv2.COLOR_RGB2GRAY)


print(x_test1.shape)

if K.image_data_format() == 'channels_first':
    print('reshape1')
    x_test1 = x_test1.reshape(1, 1, img_rows, img_cols)
    input_shape = (1, img_rows, img_cols)

else:
    print('reshape2')
    x_test1 = x_test1.reshape(1, img_rows, img_cols, 1)
    input_shape = (img_rows, img_cols, 1)

x_test1 = x_test1.astype('float32')
x_test1 /= 255
y_test1 = np.array([[ 0., 0., 1., 0., 0., 0., 0., 0., 0., 0.]])
score = model.evaluate(x_test1, y_test1, verbose=0)
print('done')
print('score of image = ')
print(score[1])
print(score[0])
score=model.predict_classes(x_test1)
print(score)

下面的代码用于加载先前训练的模型，并从先前的检查点继续训练它。如果有什么错误，请一定要提出来。

model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
                 activation='relu',
                 input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))

model.compile(loss=keras.losses.categorical_crossentropy,
              optimizer=keras.optimizers.Adadelta(),
              metrics=['accuracy'])
filepath="my_model.h5"
checkpoint = ModelCheckpoint(filepath, monitor='acc', verbose=1, save_best_only=True, mode='max')
callbacks_list = [checkpoint]

score=model.fit(x_test1, y_test1,epochs=12, batch_size=128, callbacks=callbacks_list, verbose=0)


new_model = load_model("my_model.h5")
np.testing.assert_allclose(model.predict(x_test1),
                new_model.predict(x_test1),
                1e-5)


checkpoint = ModelCheckpoint(filepath, monitor='loss', verbose=1, save_best_only=True, mode='min')
callbacks_list = [checkpoint]
new_model.fit(x_train, y_train, epochs=12, batch_size=128, callbacks=callbacks_list)

上面的代码是否会导致模型的过度拟合？如果是，我如何提高它的效率，使其能够预测任何类型的数据？需要帮助！！

Answer 1

您不应该对这些结果感到惊讶，因为您在一个数据域上训练模型，但在另一个数据域上对其进行测试。是的，如果您有一个设计更好的网络，您可能会获得稍微更好的性能，但这种差距仍然存在。

要缩小这一差距，您可以改进以下几点：

1. improve你的训练数据：(a)用更真实的数字数据训练你的模型，例如街景门牌(SVHN)数据集，或char74K数据集中的数字，以及(b)用更好的数据增强技术训练你的模型，例如，将数字样本与你的测试数据的随机背景混合:预处理测试样本，使它们看起来与训练数据中的相似，例如，对测试样本进行二值化，使其看起来类似于MNIST。在您将其提供给您的网络用于prediction.
2. improve您的模型之前，请明确考虑真实数据中的变化:例如，如果您认为测试数据中的rotation会降低您的模型性能，那么您可以在模型训练中考虑这个因素，方法是添加一个辅助任务来预测数字旋转；或者你认为一个好的模型应该只预测数字，而不是这个训练样本所属的数据集，然后你可以添加一个对抗性任务来迫使网络忘记这个information.