当前位置：首页 > wzjs >正文

广西工商网站查询企业信息seo整站优化吧

wzjs 2025/7/23 9:25:28

广西工商网站查询企业信息,seo整站优化吧,企业管理系统咨询,北京舞美设计制作公司‌一、为什么需要数据增强？ 数据增强通过对训练图像进行‌随机变换‌，能够有效： 增加数据多样性，防止过拟合提升模型对不同视角、光照条件的鲁棒性在数据量不足时显著提升模型性能二、MNIST手写数字识别实战‌ ‌1. 加载数据…

‌一、为什么需要数据增强？

数据增强通过对训练图像进行‌随机变换‌，能够有效：

增加数据多样性，防止过拟合
提升模型对不同视角、光照条件的鲁棒性
在数据量不足时显著提升模型性能

二、MNIST手写数字识别实战‌

‌1. 加载数据集与增强策略

import torch
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
import matplotlib.pyplot as plt# 高级数据增强组合
train_transform = transforms.Compose([transforms.RandomRotation(10),          # 随机旋转±10度transforms.RandomAffine(0, translate=(0.1, 0.1)), # 随机平移transforms.ColorJitter(brightness=0.2, contrast=0.2), # 调整亮度对比度transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,)) # MNIST专用标准化参数
])test_transform = transforms.Compose([transforms.ToTensor(),transforms.Normalize((0.1307,), (0.3081,))
])# 加载数据集
train_dataset = torchvision.datasets.MNIST(root='./data', train=True, download=True,transform=train_transform
)test_dataset = torchvision.datasets.MNIST(root='./data',train=False,transform=test_transform
)# 可视化增强效果
def plot_augmented_samples():fig, axes = plt.subplots(3, 5, figsize=(15, 7))for i in range(3):for j in range(5):img, label = train_dataset[i*5 + j]axes[i,j].imshow(img.squeeze(), cmap='gray')axes[i,j].axis('off')plt.suptitle("数据增强效果示例", y=0.9)plt.show()plot_augmented_samples()

‌2. 构建卷积神经网络（CNN）

class CNN(nn.Module):def __init__(self):super(CNN, self).__init__()self.conv_layers = nn.Sequential(nn.Conv2d(1, 32, 3, padding=1),  # 输入通道1，输出32，3x3卷积nn.BatchNorm2d(32),              # 批标准化nn.ReLU(),nn.MaxPool2d(2),                 # 池化层 2x2nn.Conv2d(32, 64, 3, padding=1),nn.BatchNorm2d(64),nn.ReLU(),nn.MaxPool2d(2))self.fc_layers = nn.Sequential(nn.Linear(64*7*7, 512),          # 全连接层nn.Dropout(0.5),                 # Dropout防止过拟合nn.ReLU(),nn.Linear(512, 10))def forward(self, x):x = self.conv_layers(x)x = x.view(x.size(0), -1)            # 展平特征图return self.fc_layers(x)model = CNN()
print(model)

三、高级训练技巧‌

‌1. 学习率调度器

from torch.optim.lr_scheduler import StepLRoptimizer = torch.optim.Adam(model.parameters(), lr=0.001)
scheduler = StepLR(optimizer, step_size=5, gamma=0.1)  # 每5个epoch学习率×0.1

‌2. 早停法（Early Stopping）

class EarlyStopper:def __init__(self, patience=3, min_delta=0):self.patience = patienceself.min_delta = min_deltaself.counter = 0self.min_loss = float('inf')def __call__(self, val_loss):if val_loss < self.min_loss - self.min_delta:self.min_loss = val_lossself.counter = 0else:self.counter += 1if self.counter >= self.patience:return Truereturn False

3. 完整训练流程

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)train_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=128, shuffle=True
)test_loader = torch.utils.data.DataLoader(dataset=test_dataset,batch_size=256,shuffle=False
)criterion = nn.CrossEntropyLoss()
early_stopper = EarlyStopper(patience=3)def train(epoch):model.train()total_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()total_loss += loss.item()if batch_idx % 100 == 0:print(f'Train Epoch: {epoch} [{batch_idx*len(data)}/{len(train_loader.dataset)}]'f'\tLoss: {loss.item():.4f}')return total_loss / len(train_loader)def validate():model.eval()val_loss = 0correct = 0with torch.no_grad():for data, target in test_loader:data, target = data.to(device), target.to(device)output = model(data)val_loss += criterion(output, target).item()pred = output.argmax(dim=1)correct += pred.eq(target).sum().item()val_loss /= len(test_loader)accuracy = 100. * correct / len(test_loader.dataset)print(f'\nValidation set: Average loss: {val_loss:.4f}, Accuracy: {correct}/{len(test_loader.dataset)} ({accuracy:.2f}%)\n')return val_loss, accuracy# 训练循环
best_accuracy = 0
for epoch in range(1, 20):train_loss = train(epoch)val_loss, accuracy = validate()scheduler.step()# 保存最佳模型if accuracy > best_accuracy:torch.save(model.state_dict(), "best_model.pth")best_accuracy = accuracy# 早停判断if early_stopper(val_loss):print("Early stopping triggered!")break

‌四、模型性能分析‌

‌1. 混淆矩阵可视化

from sklearn.metrics import confusion_matrix
import seaborn as snsmodel.load_state_dict(torch.load("best_model.pth"))
model.eval()all_preds = []
all_targets = []
with torch.no_grad():for data, target in test_loader:data = data.to(device)output = model(data)pred = output.argmax(dim=1)all_preds.extend(pred.cpu().numpy())all_targets.extend(target.numpy())cm = confusion_matrix(all_targets, all_preds)
plt.figure(figsize=(10,8))
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues', xticklabels=range(10), yticklabels=range(10))
plt.xlabel('Predicted')
plt.ylabel('True')
plt.title('Confusion Matrix')
plt.show()

‌2. 错误样本分析

import numpy as np# 找出错误预测的样本
errors = np.where(np.array(all_preds) != np.array(all_targets))# 可视化典型错误
plt.figure(figsize=(12,6))
for i in range(6):idx = errors[i]img = test_dataset[idx].squeeze()plt.subplot(2,3,i+1)plt.imshow(img, cmap='gray')plt.title(f"True: {all_targets[idx]}, Pred: {all_preds[idx]}")plt.axis('off')
plt.tight_layout()
plt.show()