基于卷积神经网络的手写数字识别

首先看全连接神经网络存在的问题：

• 全连接有Flatten层，把28×28的图像（在横、纵方向都有信息的图像）压平成（None，784）的一维数据，减少了信息
• 全连接神经网络层数少，但参数太大，下一层的神经元都与上一层的神经元有连接

基于卷积神经网络的手写数字识别

1、加载数据

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import SGD,Adam
%matplotlib inline

(x_train,y_train),(x_test,y_test) = tf.keras.datasets.mnist.load_data(path='mnist.npz')

# 预处理
x_train, x_test = x_train / 255.0, x_test / 255.0    # 归一到[0,1]
y_train = tf.keras.utils.to_categorical(y_train)   #转one-hot
y_test = tf.keras.utils.to_categorical(y_test)

2、搭建CNN模型

model = tf.keras.models.Sequential([
    # 二维卷积，卷积核的大小为3×3，激活函数为relu
    tf.keras.layers.Conv2D(16,(3,3), activation='relu', input_shape = (28,28,1)),
    # 最大值池化
    tf.keras.layers.MaxPooling2D((2,2)),
    
    # 二次卷积，32个卷积核
    tf.keras.layers.Conv2D(32,(3,3),activation='relu'),
    tf.keras.layers.MaxPooling2D((2,2)),
    
    # 三层卷积，16个3×3的卷积核
    tf.keras.layers.Conv2D(16,(3,3),activation='relu'),
    
    # Flatten
    tf.keras.layers.Flatten(),
    # 20 全连接层
    tf.keras.layers.Dense(20,activation='relu'),
    # 10 全连接层
    tf.keras.layers.Dense(10,activation='softmax')
])
print(model.summary())

# Model: "sequential"
# _________________________________________________________________
# Layer (type)                 Output Shape              Param #   
# =================================================================
# conv2d (Conv2D)              (None, 26, 26, 16)        160       
# _________________________________________________________________
# max_pooling2d (MaxPooling2D) (None, 13, 13, 16)        0         
# _________________________________________________________________
# conv2d_1 (Conv2D)            (None, 11, 11, 32)        4640      
# _________________________________________________________________
# max_pooling2d_1 (MaxPooling2 (None, 5, 5, 32)          0         
# _________________________________________________________________
# conv2d_2 (Conv2D)            (None, 3, 3, 16)          4624      
# _________________________________________________________________
# flatten (Flatten)            (None, 144)               0         
# _________________________________________________________________
# dense (Dense)                (None, 20)                2900      
# _________________________________________________________________
# dense_1 (Dense)              (None, 10)                210       
# =================================================================
# Total params: 12,534
# Trainable params: 12,534
# Non-trainable params: 0
# _________________________________________________________________
# None

参数计算：

• 卷积第一层（16个3×3的卷积核）：16 × 3 × 3 + 16 = 160
• 最大池化层（MaxPooling2D）：0
• 卷积第二层（32个3×3的卷积核）：3 × 3 × 16 × 32 + 32 = 4640
• 卷积第三层（16个3×3的卷积核）：3 × 3 × 16 × 32 × 16+ 16 = 4624
• Flatten层：0
• 全连接第一层：144 × 20 + 20 = 2900
• 全连接第二层：20 × 10 + 10 = 210
• 总参数量：160 + 4640 + 4624 + 2900 + 210 = 12534

print(x_train.shape)
x_train = x_train.reshape(-1,28,28,1)  # -1 代表的是图像的个数
x_train.shape
# out>>>
# (60000, 28, 28)
# (60000, 28, 28, 1)

x_test = x_test.reshape(-1,28,28,1)
x_test.shape
# out>>>   (10000, 28, 28, 1)

3、训练模型

# 设置优化器和损失函数
model.compile(
    optimizer=tf.keras.optimizers.Adam(),
    loss = tf.keras.losses.categorical_crossentropy,
    metrics=['acc']
)
history = model.fit(x_train,y_train,epochs=10)
history.history['acc']
# out>>>
# [0.9101667,
#  0.9771,
#  0.9823,
#  0.98545,
#  0.9877,
#  0.98895,
#  0.99011666,
#  0.991,
#  0.99233335,
#  0.99245]

# 可视化
plt.plot(history.history['acc'])
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.show()