网络结构:
1:输入,大小表示为(C*H*W)=(channel,hight,width),下同
输入图像大小为32*32
2:C1卷积层
输入通道1,输入通道6,卷积核大小5*5,步长1,填充2,得到输出特征图大小(6*28*28)
3:S2池化层
输入大小6*28*28,卷积核2*2,步长2,使得特征图变为原来的一半,输出大小为(6*14*14)
4:C3卷积层
输入通道6,输出通道16,输入大小(6*14*14),卷积核5*5,输出大小(16*10*10)
5:S4池化层
输入大小16*10*10
import torch
import torch.nn as nn # 引入pytorch函数库
import torchvision # 引入pytorch的图形库(torchvision.datasets/.models/.transforms/.utils)
import torchvision.transforms as transforms # 图形变换(如tensor转换)
# 是否使用GPU
DEVICE = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
# 设置超参数
NUM_CLASSES = 10 # 分类数目
EPOCHS = 5 # 训练轮次
BATCH_SIZE = 100 # 一轮训练批量大小
LR = 0.001 # 学习率learning_rate
# 创建数据集(root:设置数据集保存路径,train:是否为训练数据集,transforms:将PIL或numpy.ndarray转换为tensor格式,download:是否下载数据集)
train_dataset = torchvision.datasets.MNIST(root='data/', train=True, transform=transforms.ToTensor(), download=True)
test_dataset = torchvision.datasets.MNIST(root='data/', train=False, transform=transforms.ToTensor(), download=False)
# 加载数据集(torch.utilsdata.DataLoader:torch的数据集加载器,dataset:要加载的数据集,batch_size:批量大小,shuffle:是否打乱)
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=False)
# 构造网络模型
# input_size(1*28*28)
# 卷积:(n-k+2p)/s+1 池化:(n-k)/s+1
class LeNet(nn.Module):
def __init__(self, num_class=10):
super(LeNet, self).__init__() # 继承父类所有属性和方法,父类属性用父类的方法初始化
self.conv1 = nn.Sequential( # nn.Sequential:相当于一个容器,将一系列操作包含其中
nn.Conv2d(1, 6, 5, 1, 2), # Conv2d(in_channels,out_channels,kernel_size,stride,padding) out_size(6*28*28)
nn.ReLU(), # ReLu()激活函数引入非线性,把负值变为0,正值不变
nn.MaxPool2d(kernel_size=2, stride=2) # 最大池化层,MaxPool2d(kernel_size,stride,padding) out_size(6*14*14)
)
self.conv2 = nn.Sequential(
nn.Conv2d(6, 16, 5), # out_size(16*10*10)
nn.ReLU(),
nn.MaxPool2d(2, 2) # out_size(16*5*5)
)
self.fc1 = nn.Sequential(
nn.Linear(16 * 5 * 5, 120), # 全连接层nn.Linear(in_features,out_features)输入和输出的二维张量大小或神经元个数
nn.ReLU()
)
self.fc2 = nn.Sequential(
nn.Linear(120, 84),
nn.ReLU()
)
self.fc3 = nn.Linear(84, 10) # 最后一层得到要数字分类的10类概率值
# 前向传播
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size()[0], -1) # 表示将多维的tensor数据展平成一维,torch.view(a,b):重构成a*b维的张量 torch.view(a,-1):-1表示列需要自动计算列数
x = self.fc1(x)
x = self.fc2(x)
x = self.fc3(x)
return x
model = LeNet(NUM_CLASSES).to(DEVICE) # 将模型加载到设备上
# 构造损失函数和优化器
criterion = nn.CrossEntropyLoss() # 交叉熵损失函数
optimizer = torch.optim.Adam(model.parameters(), lr=LR)
# 训练模型
total_step = len(train_loader) # 总训练次数
for epoch in range(EPOCHS):
for i, (images, labels) in enumerate(train_loader):
images = images.to(DEVICE)
labels = labels.to(DEVICE)
# 前向传播
outputs = model(images)
loss = criterion(outputs, labels)
# 反向传播,更新优化器
optimizer.zero_grad() # 梯度置零
loss.backward() # loss反向传播计算梯度
optimizer.step() # 更新网络参数
if (i + 1) % 100 == 0:
print(f'Epoch[{epoch + 1}/{EPOCHS}], Step[{i + 1}/{total_step}], Loss:{loss.item():.4f}')
# 测试模型
model.eval() # 切换到评估模式而非训练模式即固定参数
with torch.no_grad():
correct = 0
total = 0
for images, labels in test_loader:
images = images.to(DEVICE)
labels = labels.to(DEVICE)
outputs = model(images)
_, predicted = torch.max(outputs.data, 1) # 下划线没有实际意义,表示的是具体的value,用其他变量表示也可以,1表示输出所在行的最大值,为0时表示输出所在列的最大值
total += labels.size(0) # 数据总量
correct += (predicted == labels).sum().item() # 总准确个数
print(f'Accuracy:{(100 * correct / total)}%')
# 模型保存
torch.save(model.state_dict(), 'LeNet.pth')
import torch
import torch.nn as nn
import torchvision
import matplotlib.pyplot as plt
from PIL import Image
from torchvision import transforms
import numpy as np
class LeNet(nn.Module):
def __init__(self, num_class=10):
super(LeNet, self).__init__() # 继承父类所有属性和方法,父类属性用父类的方法初始化
self.conv1 = nn.Sequential( # nn.Sequential:相当于一个容器,将一系列操作包含其中
nn.Conv2d(1, 6, 5, 1, 2), # Conv2d(in_channels,out_channels,kernel_size,stride,padding) out_size(6*28*28)
nn.ReLU(), # ReLu()激活函数引入非线性,把负值变为0,正值不变
nn.MaxPool2d(kernel_size=2, stride=2) # 最大池化层,MaxPool2d(kernel_size,stride,padding) out_size(6*14*14)
)
self.conv2 = nn.Sequential(
nn.Conv2d(6, 16, 5), # out_size(16*10*10)
nn.ReLU(),
nn.MaxPool2d(2, 2) # out_size(16*5*5)
)
self.fc1 = nn.Sequential(
nn.Linear(16 * 5 * 5, 120), # 全连接层nn.Linear(in_features,out_features)输入和输出的二维张量大小或神经元个数
nn.ReLU()
)
self.fc2 = nn.Sequential(
nn.Linear(120, 84),
nn.ReLU()
)
self.fc3 = nn.Linear(84, 10) # 最后一层得到要数字分类的10类概率值
# 前向传播
def forward(self, x):
x = self.conv1(x)
x = self.conv2(x)
x = x.view(x.size()[0], -1) # 表示将多维的tensor数据展平成一维,torch.view(a,b):重构成a*b维的张量 torch.view(a,-1):-1表示列需要自动计算列数
x = self.fc1(x)
x = self.fc2(x)
x = self.fc3(x)
return x
# 读取图片
test_img = 'newTest6.png'
img = plt.imread(test_img)
images = Image.open(test_img)
images = images.resize((28, 28))
images = images.convert('L')
transform = transforms.ToTensor()
images = transform(images)
images = images.resize(1, 1, 28, 28)
# 加载网络模型
model = LeNet()
model.load_state_dict(torch.load('LeNet.pth'))
# model = torch.load('LeNet.pth')
model.eval()
outputs = model(images)
values, indices = outputs.data.max(1) # 返回最大概率值和下标
plt.title('{}'.format((int(indices[0]))))
plt.imshow(img)
plt.show()