import torch
import torch.nn as nn# 定義通道注意力
class ChannelAttention(nn.Module):def __init__(self, in_channels, ratio=16):"""通道注意力機制初始化參數:in_channels: 輸入特征圖的通道數ratio: 降維比例，用于減少參數量，默認為16"""super().__init__()# 全局平均池化，將每個通道的特征圖壓縮為1x1，保留通道間的平均值信息self.avg_pool = nn.AdaptiveAvgPool2d(1)# 全局最大池化，將每個通道的特征圖壓縮為1x1，保留通道間的最顯著特征self.max_pool = nn.AdaptiveMaxPool2d(1)# 共享全連接層，用于學習通道間的關系# 先降維（除以ratio），再通過ReLU激活，最后升維回原始通道數self.fc = nn.Sequential(nn.Linear(in_channels, in_channels // ratio, bias=False),  # 降維層nn.ReLU(),  # 非線性激活函數nn.Linear(in_channels // ratio, in_channels, bias=False)   # 升維層)# Sigmoid函數將輸出映射到0-1之間，作為各通道的權重self.sigmoid = nn.Sigmoid()def forward(self, x):"""前向傳播函數參數:x: 輸入特征圖，形狀為 [batch_size, channels, height, width]返回:調整后的特征圖，通道權重已應用"""# 獲取輸入特征圖的維度信息，這是一種元組的解包寫法b, c, h, w = x.shape# 對平均池化結果進行處理：展平后通過全連接網絡avg_out = self.fc(self.avg_pool(x).view(b, c))# 對最大池化結果進行處理：展平后通過全連接網絡max_out = self.fc(self.max_pool(x).view(b, c))# 將平均池化和最大池化的結果相加并通過sigmoid函數得到通道權重attention = self.sigmoid(avg_out + max_out).view(b, c, 1, 1)# 將注意力權重與原始特征相乘，增強重要通道，抑制不重要通道return x * attention #這個運算是pytorch的廣播機制

## 空間注意力模塊
class SpatialAttention(nn.Module):def __init__(self, kernel_size=7):super().__init__()self.conv = nn.Conv2d(2, 1, kernel_size, padding=kernel_size//2, bias=False)self.sigmoid = nn.Sigmoid()def forward(self, x):# 通道維度池化avg_out = torch.mean(x, dim=1, keepdim=True)  # 平均池化：(B,1,H,W)max_out, _ = torch.max(x, dim=1, keepdim=True)  # 最大池化：(B,1,H,W)pool_out = torch.cat([avg_out, max_out], dim=1)  # 拼接：(B,2,H,W)attention = self.conv(pool_out)  # 卷積提取空間特征return x * self.sigmoid(attention)  # 特征與空間權重相乘

## CBAM模塊
class CBAM(nn.Module):def __init__(self, in_channels, ratio=16, kernel_size=7):super().__init__()self.channel_attn = ChannelAttention(in_channels, ratio)self.spatial_attn = SpatialAttention(kernel_size)def forward(self, x):x = self.channel_attn(x)x = self.spatial_attn(x)return x

# 測試下通過CBAM模塊的維度變化
# 輸入卷積的尺寸為
# 假設輸入特征圖：batch=2，通道=512，尺寸=26x26
x = torch.randn(2, 512, 26, 26) 
cbam = CBAM(in_channels=512)
output = cbam(x)  # 輸出形狀不變：(2, 512, 26, 26)
print(f"Output shape: {output.shape}")  # 驗證輸出維度

?cnn+CBAM

import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt
import numpy as np# 設置中文字體支持
plt.rcParams["font.family"] = ["SimHei"]
plt.rcParams['axes.unicode_minus'] = False  # 解決負號顯示問題# 檢查GPU是否可用
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"使用設備: {device}")# 數據預處理（與原代碼一致）
train_transform = transforms.Compose([transforms.RandomCrop(32, padding=4),transforms.RandomHorizontalFlip(),transforms.ColorJitter(brightness=0.2, contrast=0.2, saturation=0.2, hue=0.1),transforms.RandomRotation(15),transforms.ToTensor(),transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])test_transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])# 加載數據集（與原代碼一致）
train_dataset = datasets.CIFAR10(root='./data', train=True, download=True, transform=train_transform)
test_dataset = datasets.CIFAR10(root='./data', train=False, transform=test_transform)
train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False)

# 定義帶有CBAM的CNN模型
class CBAM_CNN(nn.Module):def __init__(self):super(CBAM_CNN, self).__init__()# ---------------------- 第一個卷積塊（帶CBAM） ----------------------self.conv1 = nn.Conv2d(3, 32, kernel_size=3, padding=1)self.bn1 = nn.BatchNorm2d(32) # 批歸一化self.relu1 = nn.ReLU()self.pool1 = nn.MaxPool2d(kernel_size=2)self.cbam1 = CBAM(in_channels=32)  # 在第一個卷積塊后添加CBAM# ---------------------- 第二個卷積塊（帶CBAM） ----------------------self.conv2 = nn.Conv2d(32, 64, kernel_size=3, padding=1)self.bn2 = nn.BatchNorm2d(64)self.relu2 = nn.ReLU()self.pool2 = nn.MaxPool2d(kernel_size=2)self.cbam2 = CBAM(in_channels=64)  # 在第二個卷積塊后添加CBAM# ---------------------- 第三個卷積塊（帶CBAM） ----------------------self.conv3 = nn.Conv2d(64, 128, kernel_size=3, padding=1)self.bn3 = nn.BatchNorm2d(128)self.relu3 = nn.ReLU()self.pool3 = nn.MaxPool2d(kernel_size=2)self.cbam3 = CBAM(in_channels=128)  # 在第三個卷積塊后添加CBAM# ---------------------- 全連接層 ----------------------self.fc1 = nn.Linear(128 * 4 * 4, 512)self.dropout = nn.Dropout(p=0.5)self.fc2 = nn.Linear(512, 10)def forward(self, x):# 第一個卷積塊x = self.conv1(x)x = self.bn1(x)x = self.relu1(x)x = self.pool1(x)x = self.cbam1(x)  # 應用CBAM# 第二個卷積塊x = self.conv2(x)x = self.bn2(x)x = self.relu2(x)x = self.pool2(x)x = self.cbam2(x)  # 應用CBAM# 第三個卷積塊x = self.conv3(x)x = self.bn3(x)x = self.relu3(x)x = self.pool3(x)x = self.cbam3(x)  # 應用CBAM# 全連接層x = x.view(-1, 128 * 4 * 4)x = self.fc1(x)x = self.relu3(x)x = self.dropout(x)x = self.fc2(x)return x# 初始化模型并移至設備
model = CBAM_CNN().to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)
scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', patience=3, factor=0.5)

# 訓練函數
def train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs):model.train()all_iter_losses = []iter_indices = []train_acc_history = []test_acc_history = []train_loss_history = []test_loss_history = []for epoch in range(epochs):running_loss = 0.0correct = 0total = 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()iter_loss = loss.item()all_iter_losses.append(iter_loss)iter_indices.append(epoch * len(train_loader) + batch_idx + 1)running_loss += iter_loss_, predicted = output.max(1)total += target.size(0)correct += predicted.eq(target).sum().item()if (batch_idx + 1) % 100 == 0:print(f'Epoch: {epoch+1}/{epochs} | Batch: {batch_idx+1}/{len(train_loader)} 'f'| 單Batch損失: {iter_loss:.4f} | 累計平均損失: {running_loss/(batch_idx+1):.4f}')epoch_train_loss = running_loss / len(train_loader)epoch_train_acc = 100. * correct / totaltrain_acc_history.append(epoch_train_acc)train_loss_history.append(epoch_train_loss)# 測試階段model.eval()test_loss = 0correct_test = 0total_test = 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()_, predicted = output.max(1)total_test += target.size(0)correct_test += predicted.eq(target).sum().item()epoch_test_loss = test_loss / len(test_loader)epoch_test_acc = 100. * correct_test / total_testtest_acc_history.append(epoch_test_acc)test_loss_history.append(epoch_test_loss)scheduler.step(epoch_test_loss)print(f'Epoch {epoch+1}/{epochs} 完成 | 訓練準確率: {epoch_train_acc:.2f}% | 測試準確率: {epoch_test_acc:.2f}%')plot_iter_losses(all_iter_losses, iter_indices)plot_epoch_metrics(train_acc_history, test_acc_history, train_loss_history, test_loss_history)return epoch_test_acc# 繪圖函數
def plot_iter_losses(losses, indices):plt.figure(figsize=(10, 4))plt.plot(indices, losses, 'b-', alpha=0.7, label='Iteration Loss')plt.xlabel('Iteration（Batch序號）')plt.ylabel('損失值')plt.title('每個 Iteration 的訓練損失')plt.legend()plt.grid(True)plt.tight_layout()plt.show()def plot_epoch_metrics(train_acc, test_acc, train_loss, test_loss):epochs = range(1, len(train_acc) + 1)plt.figure(figsize=(12, 4))plt.subplot(1, 2, 1)plt.plot(epochs, train_acc, 'b-', label='訓練準確率')plt.plot(epochs, test_acc, 'r-', label='測試準確率')plt.xlabel('Epoch')plt.ylabel('準確率 (%)')plt.title('訓練和測試準確率')plt.legend()plt.grid(True)plt.subplot(1, 2, 2)plt.plot(epochs, train_loss, 'b-', label='訓練損失')plt.plot(epochs, test_loss, 'r-', label='測試損失')plt.xlabel('Epoch')plt.ylabel('損失值')plt.title('訓練和測試損失')plt.legend()plt.grid(True)plt.tight_layout()plt.show()# 執行訓練
epochs = 50
print("開始使用帶CBAM的CNN訓練模型...")
final_accuracy = train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs)
print(f"訓練完成！最終測試準確率: {final_accuracy:.2f}%")# # 保存模型
# torch.save(model.state_dict(), 'cifar10_cbam_cnn_model.pth')
# print("模型已保存為: cifar10_cbam_cnn_model.pth")

@浙大疏錦行?