原文链接：https://data-flair.training/blogs/python-deep-learning-project-handwritten-digit-recognition/

原文讲得很详细，这里补充一些注释。由于直接从库导入mnist数据集需要的时间非常久，因此这里导入的是本地已下载好的mnist数据集。（但我怀疑我下了假的数据集，咋验证准确率这么低，所以这里不提供了）

import keras
from keras import backend as K
import numpy as np
from keras.datasets import mnist
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten
from keras.layers import Conv2D, MaxPooling2D

batch_size = 128   #一次训练所选取的样本数
num_classes = 10   #分类个数
epochs = 10        #训练轮数


#读取已下载到本地的数据集
f=np.load('C:/Users/Administrator/.keras/datasets/mnist.npz')
x_train,y_train=f['x_train'],f['y_train']
x_test,y_test=f['x_test'],f['y_test']
#print(x_train.shape, y_train.shape)


#数据预处理
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)
input_shape = (28, 28, 1)

x_train = x_train.astype('float32')  #转换数据类型
x_test = x_test.astype('float32')
x_train /= 255      #归一化
x_test /= 255 
y_train = keras.utils.to_categorical(y_train, num_classes)   #将整形数组转化为二元类型矩阵
y_test = keras.utils.to_categorical(y_test, num_classes)
#print('x_train shape:', x_train.shape)
#print(x_train.shape[0], 'train samples')
#print(x_test.shape[0], 'test samples')


#创建CNN模型
model = Sequential()  #这里采用顺序模型构建CNN
#输入层，这里指定输入数据形状为28*28*1 卷积核数量为32 形状为3*3
model.add(Conv2D(32, kernel_size=(3, 3),activation='relu',input_shape=input_shape))
#添加中间层
model.add(Conv2D(64, (3, 3), activation='relu'))   #卷积层
model.add(MaxPooling2D(pool_size=(2, 2)))          #最大池化层
model.add(Dropout(0.25))                           #通过Dropout防止过拟合
model.add(Flatten())                               #展平层
model.add(Dense(256, activation='relu'))           #全连接层
model.add(Dropout(0.5))
model.add(Dense(num_classes, activation='softmax'))
#损失函数
model.compile(loss=keras.losses.categorical_crossentropy,optimizer=keras.optimizers.Adadelta(),metrics=['accuracy'])

#训练模型
hist = model.fit(x_train, y_train,batch_size=batch_size,epochs=epochs,verbose=2,validation_data=(x_test, y_test))
print("模型训练完成")

#模型评估
score = model.evaluate(x_test, y_test, verbose=0)
print('test loss: ', score[0])
print('test accuracy: ', score[1])