Deep Learning 100 Examples - Convolutional Neural Network (CNN) for mnist handwritten digit recognition

活动地址：CSDN21天学习挑战赛

深度学习100例 —— 卷积神经网络(CNN)实现mnist手写数字识别

K老师用的tensorflow框架,啊这,Create a new virtual environment.

Python：3.7

编译器：jupyter notebook

深度学习框架：TensorFlow2.7.0

1. 前期准备工作

1.1 设置GPU

Here I finally chose to use itAutoDL的算力,I hate to run on my computer.

import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU") # tf.config.list_physical_devices# Get a specific computing device type on the current host（如 GPU 或 CPU ）的列表

if gpus:
    gpu0 = gpus[0] #如果有多个GPU,仅使用第0个GPU
    tf.config.experimental.set_memory_growth(gpu0, True) #设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpu0],"GPU") # Set the list of visible devices

1.2 导入数据

import tensorflow as tf
from tensorflow.keras import datasets, layers, models
import matplotlib.pyplot as plt

(train_images, train_labels), (test_images, test_labels) = datasets.mnist.load_data()

MNISTThe handwritten digits dataset is from the National Institute of Standards and Technology,is one of the well-known public datasets.The digital pictures in the dataset are made by 250个不同职业的人纯手写绘制.

MNISTIncluded in the handwritten digits dataset70000张图片,其中60000Zhang is the training data,10000Zhang is the test data,70000pictures are28*28

28 x 28 Convert pixels to vectors,得到长度为28 x 28 = 784 的向量,

1.3 数据归一化处理

作用：

• 使不同量纲的特征处于同一数值量级,减少方差大的特征的影响,使模型更准确.
• 加快学习算法的收敛速度.

归一化：把数变为(0 , 1)之间的小数;Scaling only follows the maximum value、最小值的差别有关.

# 将像素的值标准化至0到1的区间内.
train_images, test_images = train_images / 255.0, test_images / 255.0

train_images.shape,test_images.shape,train_labels.shape,test_labels.shape

1.4 可视化图片

plt.figure(figsize=(20,10)) # 设置绘图窗口大小
for i in range(20):
    plt.subplot(5,10,i+1) # 整个figure被分为5行10列,从左到右,从上到下编号为1,2,3...
    plt.xticks([]) # 绘制x轴标签的方向
    plt.yticks([]) # 绘制y轴标签的方向
    plt.grid(False) # Turn off gridline display
    plt.imshow(train_images[i], cmap=plt.cm.binary) # cmap=plt.cm.binary → 显示黑白图像,展示0-19的图像
    plt.xlabel(train_labels[i]) # 设置xThe axis labels are the corresponding label values
plt.show()

1.5 调整图片格式

#调整数据到我们需要的格式
train_images = train_images.reshape((60000, 28, 28, 1))
test_images = test_images.reshape((10000, 28, 28, 1))

train_images.shape,test_images.shape,train_labels.shape,test_labels.shape

→ 数据维度转换,增加通道维度

2. 构建CNN网络模型

2.1 构建网络

model = models.Sequential(
    [
        layers.Conv2D(32 , (3 , 3) , activation = 'relu', input_shape = [28 , 28 , 1]), # 卷积层1,卷积核3 x 3
        layers.MaxPooling2D((2, 2)), # 池化层1 , 2 x 2 采样
        layers.Conv2D(64 , (3 , 3) , activation = 'relu'), # 卷积层2 , 卷积核3 x 3
        layers.MaxPooling2D((2, 2)), # 池化层2, 2 x 2 采样

        layers.Flatten(), # Flatten层,连接卷积层与全连接层
        layers.Dense(64 , activation = 'relu'), # 全连接层,特征进一步提取
        layers.Dense(10) # 输出层,输出结果
    ]
)
model.summary()

网络结构

2.2 模型结构说明：

各层的作用：

• 输入层：Feed the data into the training network
• 卷积层：Extract features from images using convolution kernels
• 池化层：进行下采样,Represent image features with a higher level of abstraction
• Flatten层：将多维的输入一维化,Commonly used in the transition from convolutional layers to fully connected layers
• 全连接层：特征提取器
• 输出层：输出结果

3. 编译模型

model.compile(optimizer='adam',
         loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])

compile()method is used to set training time,使用的优化器optimizer、损失函数loss、准确率评测标准metrics

SparseCategoricalCrossentropy → 交叉熵损失函数,当from_logits参数为True时,会使用softmax将预测y转换为概率.

4. 训练模型

history = model.fit(train_images, train_labels, epochs=10, 
                    validation_data=(test_images, test_labels))

epoch : 迭代次数,All samples will be trained loop10次

validation_data：指定验证数据.

The accuracy of the compiled model settings,Here is the output accuracy

5. 预测

Take a look at the test set0号图片

OK,是个7

使用训练好的模型进行预测：