【Pytorch】卷积神经网络实现手写数字识别

1 加载数据

分别构建训练集和测试集（验证集）
DataLoader来迭代取数据

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torchvision import datasets,transforms 
import matplotlib.pyplot as plt
import numpy as np
%matplotlib inline


# 定义超参数 
input_size = 28  #图像的总尺寸28*28
num_classes = 10  #标签的种类数
num_epochs = 3  #训练的总循环周期
batch_size = 64  #一个撮（批次）的大小，64张图片

# 训练集
train_dataset = datasets.MNIST(root='./data',  
                            train=True,   
                            transform=transforms.ToTensor(),  
                            download=True) 

# 测试集
test_dataset = datasets.MNIST(root='./data', 
                           train=False, 
                           transform=transforms.ToTensor())

# 构建batch数据
train_loader = torch.utils.data.DataLoader(dataset=train_dataset, 
                                           batch_size=batch_size, 
                                           shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset, 
                                           batch_size=batch_size, 
                                           shuffle=True)

2 模型构建

构建卷积神经网络，一般卷积层，relu层，池化层可以写成一个套餐

class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.conv1 = nn.Sequential(         # 输入大小 (1, 28, 28)
            nn.Conv2d(
                in_channels=1,              # 灰度图
                out_channels=16,            # 要得到几多少个特征图
                kernel_size=5,              # 卷积核大小
                stride=1,                   # 步长
                padding=2,                  # 
            ),                              # 输出的特征图为 (16, 28, 28)
            nn.ReLU(),                      # relu层
            nn.MaxPool2d(kernel_size=2),    # 进行池化操作（2x2 区域）, 输出结果为： (16, 14, 14)
        )
        self.conv2 = nn.Sequential(         # 下一个套餐的输入 (16, 14, 14)
            nn.Conv2d(16, 32, 5, 1, 2),     # 输出 (32, 14, 14)
            nn.ReLU(),                      # relu层
            nn.MaxPool2d(2),                # 输出 (32, 7, 7)
        )
        self.out = nn.Linear(32 * 7 * 7, 10)   # 全连接层得到的结果

    def forward(self, x):
       
        x = self.conv1(x)
        x = self.conv2(x)
        x = x.view(x.size(0), -1)           # flatten操作，结果为：(batch_size, 32 * 7 * 7)
        output = self.out(x)
        return output

3 训练模型

准确率作为评估标准

def accuracy(predictions, labels):
    pred = torch.max(predictions.data, 1)[1] 
    rights = pred.eq(labels.data.view_as(pred)).sum() 
    return rights, len(labels) 

# 实例化
net = CNN() 
#损失函数
criterion = nn.CrossEntropyLoss() 
#优化器
optimizer = optim.Adam(net.parameters(), lr=0.001) #定义优化器，普通的随机梯度下降算法

#开始训练循环
for epoch in range(num_epochs):
    #当前epoch的结果保存下来
    train_rights = [] 
    
    for batch_idx, (data, target) in enumerate(train_loader):  #针对容器中的每一个批进行循环
        net.train()                             
        output = net(data) 
      
        loss = criterion(output, target) 
        optimizer.zero_grad() # 梯度归o
        loss.backward() 
        optimizer.step()  # 更新优化器的学习率
        right = accuracy(output, target) 
        train_rights.append(right) 

    
        if batch_idx % 100 == 0: 
            
            net.eval() 
            val_rights = [] 
            
            for (data, target) in test_loader:
                output = net(data) 
                right = accuracy(output, target) 
                val_rights.append(right)
                
            #准确率计算
            train_r = (sum([tup[0] for tup in train_rights]), sum([tup[1] for tup in train_rights]))
            val_r = (sum([tup[0] for tup in val_rights]), sum([tup[1] for tup in val_rights]))

            print('当前epoch: {
    
    } [{
    
    }/{
    
    } ({
    
    :.0f}%)]\t损失: {
    
    :.6f}\t训练集准确率: {
    
    :.2f}%\t测试集正确率: {
    
    :.2f}%'.format(
                epoch, batch_idx * batch_size, len(train_loader.dataset),
                100. * batch_idx / len(train_loader), 
                loss.data, 
                100. * train_r[0].numpy() / train_r[1], 
                100. * val_r[0].numpy() / val_r[1]))

输出

当前epoch: 0 [0/60000 (0%)]	损失: 2.287790	训练集准确率: 14.06%	测试集正确率: 11.06%
当前epoch: 0 [6400/60000 (11%)]	损失: 0.402259	训练集准确率: 75.60%	测试集正确率: 92.08%
当前epoch: 0 [12800/60000 (21%)]	损失: 0.071038	训练集准确率: 84.53%	测试集正确率: 94.49%
当前epoch: 0 [19200/60000 (32%)]	损失: 0.055919	训练集准确率: 88.09%	测试集正确率: 96.30%
当前epoch: 0 [25600/60000 (43%)]	损失: 0.065247	训练集准确率: 90.17%	测试集正确率: 97.37%
当前epoch: 0 [32000/60000 (53%)]	损失: 0.101428	训练集准确率: 91.52%	测试集正确率: 97.45%
当前epoch: 0 [38400/60000 (64%)]	损失: 0.119459	训练集准确率: 92.52%	测试集正确率: 97.69%
当前epoch: 0 [44800/60000 (75%)]	损失: 0.062872	训练集准确率: 93.21%	测试集正确率: 97.87%
当前epoch: 0 [51200/60000 (85%)]	损失: 0.044197	训练集准确率: 93.75%	测试集正确率: 97.86%
当前epoch: 0 [57600/60000 (96%)]	损失: 0.140018	训练集准确率: 94.13%	测试集正确率: 98.00%
当前epoch: 1 [0/60000 (0%)]	损失: 0.020221	训练集准确率: 100.00%	测试集正确率: 98.44%
当前epoch: 1 [6400/60000 (11%)]	损失: 0.084976	训练集准确率: 98.02%	测试集正确率: 98.33%
当前epoch: 1 [12800/60000 (21%)]	损失: 0.098251	训练集准确率: 97.92%	测试集正确率: 98.39%
当前epoch: 1 [19200/60000 (32%)]	损失: 0.078864	训练集准确率: 98.00%	测试集正确率: 98.47%
当前epoch: 1 [25600/60000 (43%)]	损失: 0.025394	训练集准确率: 98.13%	测试集正确率: 98.40%
当前epoch: 1 [32000/60000 (53%)]	损失: 0.042705	训练集准确率: 98.13%	测试集正确率: 98.28%
当前epoch: 1 [38400/60000 (64%)]	损失: 0.027868	训练集准确率: 98.13%	测试集正确率: 98.57%
当前epoch: 1 [44800/60000 (75%)]	损失: 0.010066	训练集准确率: 98.17%	测试集正确率: 98.57%
当前epoch: 1 [51200/60000 (85%)]	损失: 0.035174	训练集准确率: 98.19%	测试集正确率: 98.68%
当前epoch: 1 [57600/60000 (96%)]	损失: 0.021053	训练集准确率: 98.25%	测试集正确率: 98.61%
当前epoch: 2 [0/60000 (0%)]	损失: 0.004226	训练集准确率: 100.00%	测试集正确率: 98.46%
当前epoch: 2 [6400/60000 (11%)]	损失: 0.012750	训练集准确率: 98.69%	测试集正确率: 98.78%
当前epoch: 2 [12800/60000 (21%)]	损失: 0.071001	训练集准确率: 98.59%	测试集正确率: 98.24%
当前epoch: 2 [19200/60000 (32%)]	损失: 0.116683	训练集准确率: 98.67%	测试集正确率: 98.75%
当前epoch: 2 [25600/60000 (43%)]	损失: 0.082070	训练集准确率: 98.65%	测试集正确率: 98.79%
当前epoch: 2 [32000/60000 (53%)]	损失: 0.011719	训练集准确率: 98.65%	测试集正确率: 98.93%
当前epoch: 2 [38400/60000 (64%)]	损失: 0.044769	训练集准确率: 98.66%	测试集正确率: 98.81%
当前epoch: 2 [44800/60000 (75%)]	损失: 0.181679	训练集准确率: 98.67%	测试集正确率: 99.07%
当前epoch: 2 [51200/60000 (85%)]	损失: 0.022912	训练集准确率: 98.67%	测试集正确率: 98.77%
当前epoch: 2 [57600/60000 (96%)]	损失: 0.084802	训练集准确率: 98.69%	测试集正确率: 98.77%

4 模型保存

# 只保存模型参数
# torch.save(net.state_dict(), 'cov.pkl')
# 加载
# model = CNN() 
# model.load_state_dict(torch.load('\cov.pkl'))


# 保存
torch.save(net, 'cov.pkl')
# 加载
#model = torch.load('\cov.pkl')

5 模型加载和使用

model = torch.load('cov.pkl')
print(model)

输出

CNN(
  (conv1): Sequential(
    (0): Conv2d(1, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (1): ReLU()
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv2): Sequential(
    (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (1): ReLU()
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (out): Linear(in_features=1568, out_features=10, bias=True)
)

---

import cv2
import matplotlib.pyplot as plt

# 第一步：读取图片
img = cv2.imread('./data/test/4.png') 
print(img.shape)

# 第二步：将图片转为灰度图
img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print(img.shape)
plt.imshow(img,cmap='Greys')

# 第三步：将图片的底色和字的颜色取反
img = cv2.bitwise_not(img)
plt.imshow(img,cmap='Greys')


# 第四步：将底变成纯白色，将字变成纯黑色
img[img<=144]=0
img[img>140]=255  # 130

# 显示图片
plt.imshow(img,cmap='Greys')
 

# 第五步：将图片尺寸缩放为输入规定尺寸
img = cv2.resize(img,(28,28))

# 第六步：将数据类型转为float32
img = img.astype('float32')

# 第七步：数据正则化
img /= 255

# 第八步：增加维度为输入的规定格式
img = img.reshape(1,1, 28, 28)
print(img.shape)

# 第九步：预测
pred = model(torch.from_numpy(img))

# 第十步：输出结果
print(pred.argmax())

输出

(384, 317, 3)
(384, 317)
(1, 1, 28, 28)
tensor(4)