当前位置：首页 > ai >正文

使用PyTorch实现简单图像识别（基于MNIST手写数字数据集）的完整代码示例，包含数据加载、模型定义、训练和预测全流程

ai 2025/7/6 12:36:55

以下是一个使用PyTorch实现简单图像识别（基于MNIST手写数字数据集）的完整代码示例，包含数据加载、模型定义、训练和预测全流程：

import torch
import torch.nn as nn
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt# 设置设备
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")# ----------------------
# 1. 数据准备
# ----------------------
transform = transforms.Compose([transforms.RandomRotation(10),       # 数据增强：随机旋转transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))  # MNIST的均值和标准差
])# 加载数据集
train_set = torchvision.datasets.MNIST(root='./data',train=True,download=True,transform=transform
)test_set = torchvision.datasets.MNIST(root='./data',train=False,download=True,transform=transform
)# 创建数据加载器
train_loader = DataLoader(train_set, batch_size=64, shuffle=True)
test_loader = DataLoader(test_set, batch_size=1000, shuffle=False)# ----------------------
# 2. 构建神经网络模型
# ----------------------
class SimpleCNN(nn.Module):def __init__(self):super(SimpleCNN, self).__init__()self.conv1 = nn.Conv2d(1, 32, 3, 1)  # 输入通道1，输出通道32，卷积核3x3self.conv2 = nn.Conv2d(32, 64, 3, 1)self.dropout = nn.Dropout2d(0.25)self.fc1 = nn.Linear(9216, 128)      # 全连接层self.fc2 = nn.Linear(128, 10)        # 输出层（10个数字类别）def forward(self, x):x = torch.relu(self.conv1(x))       # 28x28 → 26x26x = torch.max_pool2d(x, 2)          # 26x26 → 13x13x = torch.relu(self.conv2(x))       # 13x13 → 11x11x = torch.max_pool2d(x, 2)          # 11x11 → 5x5x = self.dropout(x)x = torch.flatten(x, 1)            # 展平x = torch.relu(self.fc1(x))x = self.dropout(x)x = self.fc2(x)return xmodel = SimpleCNN().to(device)# ----------------------
# 3. 训练配置
# ----------------------
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)# ----------------------
# 4. 训练循环
# ----------------------
def train(epoch):model.train()train_loss = 0for batch_idx, (data, target) in enumerate(train_loader):data, target = data.to(device), target.to(device)optimizer.zero_grad()output = model(data)loss = criterion(output, target)loss.backward()optimizer.step()train_loss += loss.item()if batch_idx % 100 == 0:print(f"Train Epoch: {epoch} [{batch_idx * len(data)}/{len(train_loader.dataset)} "f"({100. * batch_idx / len(train_loader):.0f}%)]\tLoss: {loss.item():.6f}")avg_loss = train_loss / len(train_loader)return avg_loss# ----------------------
# 5. 测试函数
# ----------------------
def test():model.eval()test_loss = 0correct = 0with torch.no_grad():for data, target in test_loader:data, target = data.to(device), target.to(device)output = model(data)test_loss += criterion(output, target).item()pred = output.argmax(dim=1, keepdim=True)correct += pred.eq(target.view_as(pred)).sum().item()test_loss /= len(test_loader)accuracy = 100. * correct / len(test_loader.dataset)print(f"\nTest set: Average loss: {test_loss:.4f}, Accuracy: {correct}/{len(test_loader.dataset)} "f"({accuracy:.2f}%)\n")return accuracy# ----------------------
# 6. 执行训练
# ----------------------
epochs = 5
train_losses = []
test_accuracies = []for epoch in range(1, epochs + 1):loss = train(epoch)accuracy = test()train_losses.append(loss)test_accuracies.append(accuracy)# ----------------------
# 7. 可视化结果
# ----------------------
plt.figure(figsize=(12, 5))
plt.subplot(1, 2, 1)
plt.plot(train_losses, label='Training Loss')
plt.title('Training Loss')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()plt.subplot(1, 2, 2)
plt.plot(test_accuracies, label='Test Accuracy')
plt.title('Test Accuracy')
plt.xlabel('Epoch')
plt.ylabel('Accuracy (%)')
plt.legend()
plt.show()# ----------------------
# 8. 保存和加载模型
# ----------------------
torch.save(model.state_dict(), "mnist_cnn.pth")# 加载模型示例：
# loaded_model = SimpleCNN().to(device)
# loaded_model.load_state_dict(torch.load("mnist_cnn.pth"))# ----------------------
# 9. 单张图片预测
# ----------------------
def predict_image(img):model.eval()img = img.unsqueeze(0).to(device)  # 添加batch维度with torch.no_grad():output = model(img)_, predicted = torch.max(output.data, 1)return predicted.item()# 测试单张图片
sample_data, sample_label = next(iter(test_loader))
sample_img = sample_data[0].cpu()
prediction = predict_image(sample_img)plt.imshow(sample_img.squeeze(), cmap='gray')
plt.title(f"True: {sample_label[0]}, Predicted: {prediction}")
plt.show()

代码说明：

数据准备：

使用MNIST手写数字数据集（0-9共10类）
应用数据增强（随机旋转）和标准化处理
创建训练集和测试集的DataLoader

模型结构：

简单CNN架构：2个卷积层 + 2个全连接层
使用ReLU激活函数和MaxPooling
添加Dropout防止过拟合

训练配置：

交叉熵损失函数
Adam优化器
5个训练周期（可调整）

可视化：

绘制训练损失曲线
绘制测试准确率曲线

扩展功能：

模型保存与加载
单张图片预测演示

运行结果示例：

Using device: cuda
Train Epoch: 1 [0/60000 (0%)]  Loss: 2.307364
Train Epoch: 1 [6400/60000 (11%)] Loss: 0.362455
...
Test set: Average loss: 0.0512, Accuracy: 9853/10000 (98.53%)