深度学习笔记_6经典预训练网络LeNet-18解决FashionMNIST数据集

1、 调用模型库，定义参数，做数据预处理

import numpy as np
import torch
from torchvision.datasets import FashionMNIST
import torchvision.transforms as transforms
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torch.optim as optim
from torch import nn
from sklearn.metrics import confusion_matrix, accuracy_score, precision_score, recall_score, f1_score, roc_curve, auc
import matplotlib.pyplot as plt
from torchvision import models

# 检查 GPU 可用性
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print("Using device:", device)

# 设置超参数
train_batch_size = 64
test_batch_size = 64
learning_rate = 0.001
num_epochs = 50

# 定义数据转换操作
transform = transforms.Compose([
    transforms.RandomRotation(degrees=[-30, 30]),   # 随机旋转
    transforms.RandomHorizontalFlip(),   # 随机水平翻转
    transforms.Resize((224, 224)),  # 调整图像大小
    transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.2),   # 颜色抖动
    transforms.ToTensor(),  # 将图像转换为张量
    transforms.Normalize((0.5,), (0.5,))
])

2、下载FashionMNIST训练集

# 下载FashionMNIST训练集
trainset = FashionMNIST(root='data', train=True,
                        download=True, transform=transform)

# 下载FashionMNIST测试集
testset = FashionMNIST(root='data', train=False,
                       download=True, transform=transform)

# 创建 DataLoader 对象
train_loader = DataLoader(trainset, batch_size=train_batch_size, shuffle=True)
test_loader = DataLoader(testset, batch_size=test_batch_size, shuffle=False)

3、使用预训练的ResNet-18模型

# 使用预训练的ResNet-18模型
model = models.resnet18(pretrained=True)
# 修改最后一层，使其适应FashionMNIST的输出类别数
model.conv1 = nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3, bias=False)
model.fc = nn.Linear(model.fc.in_features, 10)
model = model.to(device)

# 冻结预训练模型的参数
for param in model.parameters():
    param.requires_grad = False

# 只训练模型的最后一层
for param in model.fc.parameters():
    param.requires_grad = True
# 初始化优化器和损失函数
optimizer = optim.Adam(model.fc.parameters(), lr=learning_rate)
criterion = nn.CrossEntropyLoss()

4、 训练循环

# 记录训练和测试过程中的损失和准确率
train_losses = []
test_losses = []
train_accuracies = []
test_accuracies = []
conf_matrix_list = []
accuracy_list = []
error_rate_list = []
precision_list = []
recall_list = []
f1_score_list = []
roc_auc_list = []

# 训练循环
for epoch in range(num_epochs):
    model.train()
    train_loss = 0.0
    correct = 0
    total = 0

    for batch_idx, (data, target) in enumerate(train_loader):
        optimizer.zero_grad()
        data, target = data.to(device), target.to(device)  # 将数据移到 GPU 上
        output = model(data)
        loss = criterion(output, target)
        loss.backward()
        optimizer.step()
        train_loss += loss.item()

        # 计算训练准确率
        _, predicted = output.max(1)
        total += target.size(0)
        correct += predicted.eq(target).sum().item()

    # 计算平均训练损失和训练准确率
    train_loss /= len(train_loader)
    train_accuracy = 100. * correct / total
    train_losses.append(train_loss)
    train_accuracies.append(train_accuracy)

    # 测试模型
    model.eval()
    test_loss = 0.0
    correct = 0
    all_labels = []
    all_preds = []

    with torch.no_grad():
        for data, target in test_loader:
            data, target = data.to(device), target.to(device)  # 将数据移到 GPU 上
            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()
            all_labels.extend(target.cpu().numpy())  # 将结果移到 CPU 上
            all_preds.extend(pred.cpu().numpy())  # 将结果移到 CPU 上

    # 计算平均测试损失和测试准确率
    test_loss /= len(test_loader)
    test_accuracy = 100. * correct / len(test_loader.dataset)
    test_losses.append(test_loss)
    test_accuracies.append(test_accuracy)

    # 计算额外的指标
    conf_matrix = confusion_matrix(all_labels, all_preds)
    conf_matrix_list.append(conf_matrix)

    accuracy = accuracy_score(all_labels, all_preds)
    accuracy_list.append(accuracy)

    error_rate = 1 - accuracy
    error_rate_list.append(error_rate)

    precision = precision_score(all_labels, all_preds, average='weighted')
    recall = recall_score(all_labels, all_preds, average='weighted')
    f1 = f1_score(all_labels, all_preds, average='weighted')

    precision_list.append(precision)
    recall_list.append(recall)
    f1_score_list.append(f1)

    fpr, tpr, thresholds = roc_curve(all_labels, all_preds, pos_label=1)
    roc_auc = auc(fpr, tpr)
    roc_auc_list.append(roc_auc)

    # 打印每个 epoch 的指标
    print(f'Epoch [{epoch + 1}/{num_epochs}] -> Train Loss: {train_loss:.4f}, Train Accuracy: {train_accuracy:.2f}%, Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.2f}%')

5、绘制Loss、Accuracy曲线图， 计算混淆矩阵

import seaborn as sns
# 绘制Loss曲线图
plt.figure()
plt.plot(train_losses, label='Train Loss', color='blue')
plt.plot(test_losses, label='Test Loss', color='red')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend()
plt.title('Loss Curve')
plt.grid(True)
plt.show()

# 绘制Accuracy曲线图
plt.figure()
plt.plot(train_accuracies, label='Train Accuracy', color='red')  # 绘制训练准确率曲线
plt.plot(test_accuracies, label='Test Accuracy', color='green')
plt.xlabel('Epoch')
plt.ylabel('Accuracy')
plt.legend()
plt.title('Accuracy Curve')
plt.grid(True)
plt.show()

# 计算混淆矩阵
confusion_mat = confusion_matrix(all_labels, all_preds)
class_labels = [str(i) for i in range(10)]
plt.figure()
sns.heatmap(confusion_mat, annot=True, fmt='d', cmap='Blues', cbar=False)
plt.xlabel('Predicted Labels')
plt.ylabel('True Labels')
plt.title('Confusion Matrix')
plt.savefig('confusion_matrix.png')
plt.show()