当前位置：首页 > news >正文

基于PyTorch利用CNN实现MNIST的手写数字识别

news 2025/8/1 12:00:49

GitHub地址：
https://github.com/gao7025/pytorch_cnn_mnist.git

1.定义模型、损失函数、优化器

定义一个卷积神经网络的网络结构，并设置一个优化器（optimizer）和一个损失准则（losscriterion）。创建一个随机梯度下降（stochasticgradientdescent）优化器

# 仅定义模型
class Net(nn.Module):def __init__(self):super(Net, self).__init__()self.conv1 = nn.Conv2d(1, 16, kernel_size=4)self.conv2 = nn.Conv2d(16, 32, kernel_size=4)self.fc1 = nn.Linear(32 * 4 * 4, 32)self.fc2 = nn.Linear(32, 10)def forward(self, x):x = f.relu(self.conv1(x))x = f.max_pool2d(x, 2)x = f.relu(self.conv2(x))x = f.max_pool2d(x, 2)x = x.view(-1, 32 * 4 * 4)x = f.relu(self.fc1(x))x = self.fc2(x)return x

# 定义模型、损失函数、优化器
from model_class import Net
model = Net()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

2.加载已有或下载mnist数据集

# 数据加载与预处理
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))  # MNIST数据集的均值和标准差
])train_dataset = datasets.MNIST(root='mnist_data', train=True, download=False, transform=transform)
test_dataset = datasets.MNIST(root='mnist_data', train=False, download=False, transform=transform)train_loader = DataLoader(train_dataset, batch_size=512, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=512)

3.模型训练

# 模型训练
def train(model, train_loader, criterion, optimizer, epoch, train_losses):model.train()train_loss = 0for batch_idx, (data, target) in enumerate(train_loader):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'Epoch: {epoch + 1}, Batch: {batch_idx}, Loss: {loss.item():.4f}')avg_train_loss = train_loss / len(train_loader)train_losses.append(avg_train_loss)return avg_train_loss

4.模型的验证与测试

# 模型验证
def evaluate(model, test_loader, criterion, test_losses, accuracies):model.eval()test_loss = 0correct = 0# 禁用梯度计算区域with torch.no_grad():for data, target in test_loader: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()avg_test_loss = test_loss / len(test_loader)accuracy = 100. * correct / len(test_loader.dataset)test_losses.append(avg_test_loss)accuracies.append(accuracy)print(f'\nTest set: Average loss: {avg_test_loss:.4f}, 'f'Accuracy: {correct}/{len(test_loader.dataset)} ({accuracy:.2f}%)\n')return avg_test_loss, accuracy

5.绘制训练与测试的loss曲线和acc曲线并调优

def plot_loss_acc(train_losses, test_losses, accuracies):"""针对训练过程绘制loss曲线和acc曲线Parameters----------train_losses : list_liketrain loss list。test_losses : list_liketest loss list。accuracies : list_likeaccuracies list。"""plt.figure(figsize=(10, 4))plt.subplot(1, 2, 1)plt.plot(train_losses, 'o-', label='Training Loss')plt.plot(test_losses, 's-', label='Test Loss')plt.xlabel('Epoch')plt.ylabel('Loss')plt.title('Training and Test Loss')plt.legend()# plt.grid(True)plt.subplot(1, 2, 2)plt.plot(accuracies, 'd-', color='green')plt.xlabel('Epoch')plt.ylabel('Accuracy (%)')plt.title('Test Accuracy')# plt.grid(True)plt.tight_layout()timestamp = str(dt.now().strftime("%y%m%d%H%M%S"))plt.savefig('./results/training_metrics_{time}.png'.format(time=timestamp))plt.show()

6.模型预测与结果可视化

加载训练好的模型和参数，并将模型设置为评估模式

from model_class import Net
model = Net()
model.load_state_dict(torch.load('mnist_cnn_model_params.pth'))
model.eval()

加载测试集数据

# 测试集数据加载
transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))  # MNIST数据集的均值和标准差
])
test_dataset = datasets.MNIST(root='mnist_data', train=False, download=False, transform=transform)
test_loader = DataLoader(test_dataset, batch_size=512)

预测结果，并将部分结果可视化展示

# 预测函数
def predict_random_samples(model, test_dataset, num_samples=9, max_show_num=16):"""随机选择样本进行预测并可视化结果"""model.eval()indices = np.random.choice(len(test_dataset), num_samples, replace=False)# 计算子图网格的行数和列数cols = int(np.ceil(np.sqrt(min(num_samples, max_show_num))))  # 向上取整rows = int(np.ceil(min(num_samples, max_show_num) / cols))# 创建相应大小的子图网格fig, axes = plt.subplots(rows, cols, figsize=(cols * 2, rows * 2))# 如果只有一个子图，axes会是一个单独的对象，需要将其转为数组if num_samples == 1:axes = np.array([axes])else:# 将多维数组展平为一维数组axes = axes.flatten()# fig, axes = plt.subplots(3, 3, figsize=(10, 6))# axes = axes.flatten()correct = 0with torch.no_grad():for i, idx in enumerate(indices):image, true_label = test_dataset[idx]output = model(image.unsqueeze(0))pred = output.argmax(dim=1)correct += pred.eq(true_label).sum().item()# 反标准化以便正确显示图像img = image.squeeze().numpy() * 0.3081 + 0.1307# 只处理有效的子图索引if i < len(axes) and i <= max_show_num:axes[i].imshow(img, cmap='gray')axes[i].set_title(f'Pred: {pred}, True: {true_label}',color=('green' if pred == true_label else 'red'))axes[i].axis('off')# 隐藏多余的子图for i in range(num_samples, len(axes)):axes[i].axis('off')accuracy = 100. * correct / num_samplesprint(f'Accuracy: {correct}/{num_samples} ({accuracy:.2f}%)\n')plt.tight_layout()timestamp = str(dt.now().strftime("%y%m%d%H%M%S"))plt.savefig('./results/prediction_samples_{time}.png'.format(time=timestamp))plt.show()print(f"预测样本已保存为 'prediction_samples_{timestamp}.png'")