基于OpenMAX的细粒度分类模型：拒绝未知类别的Python实现

本项目使用两层ResNet作为特征提取器，并利用LDA线性判别损失来最大化类间分离和最小化类内分离。为了拒绝未知类别，我们使用了OpenMAX分类器，并采用交叉熵损失函数进行训练。使用CICIDS2017数据集进行测试，该模型能有效地进行细粒度分类并拒绝未知类别。

模型结构

特征提取器: 两层ResNet
损失函数: LDA线性判别损失
分类器: OpenMAX分类器
优化器: Adam
学习率调整策略: StepLR

数据集

使用CICIDS2017数据集进行测试。

已知类: 数据集中样本前四最多的类型
未知类: 数据集中剩余类别

训练过程

导入必要的库

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import torchvision
from torchvision import datasets, models, transforms
import numpy as np
import matplotlib.pyplot as plt
import time
import os
import copy

设置训练参数

# 设置训练集和验证集的路径
data_dir = 'path_to_dataset'
# 设置模型保存路径
save_dir = 'path_to_save_model'

# 设置训练参数
num_epochs = 10
batch_size = 32
learning_rate = 0.001

设置模型特征提取器

pretrained_model = models.resnet50(pretrained=True)
pretrained_model = nn.Sequential(*list(pretrained_model.children())[:-1])

设置LDA线性判别损失

class LDA_Loss(nn.Module):
    def __init__(self, num_classes):
        super(LDA_Loss, self).__init__()
        self.num_classes = num_classes
        
    def forward(self, features, labels):
        mean_features = torch.zeros((self.num_classes, features.size(1))).cuda()
        class_count = torch.zeros(self.num_classes).cuda()
        
        for i in range(self.num_classes):
            class_features = features[labels == i]
            mean_features[i] = torch.mean(class_features, dim=0)
            class_count[i] = class_features.size(0)
        
        total_mean = torch.mean(mean_features, dim=0)
        between_class_var = torch.sum(class_count * (mean_features - total_mean) ** 2)
        within_class_var = torch.sum((features - mean_features[labels]) ** 2)
        
        lda_loss = between_class_var / within_class_var
        return lda_loss

设置openMAX分类器

class OpenMAX(nn.Module):
    def __init__(self, num_classes):
        super(OpenMAX, self).__init__()
        self.num_classes = num_classes
        self.softmax = nn.Softmax(dim=1)
        self.alpha = nn.Parameter(torch.zeros(self.num_classes))
        self.beta = nn.Parameter(torch.ones(self.num_classes))
        
    def forward(self, logits, features):
        prob = self.softmax(logits)
        openmax_prob = torch.zeros_like(prob)
        
        for i in range(logits.size(0)):
            openmax_prob[i] = self.softmax(self.alpha + self.beta * torch.log(prob[i]))
        
        return openmax_prob

设置交叉熵损失函数

criterion = nn.CrossEntropyLoss()

设置数据预处理和加载器

data_transforms = {
    'train': transforms.Compose([
        transforms.RandomResizedCrop(224),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
    'val': transforms.Compose([
        transforms.Resize(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
}

image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x]) for x in ['train', 'val']}
dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=batch_size, shuffle=True, num_workers=4) for x in ['train', 'val']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
class_names = image_datasets['train'].classes
num_classes = len(class_names)

设置GPU加速

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

定义训练和验证函数

def train_model(model, criterion, optimizer, scheduler, num_epochs):
    since = time.time()

    best_model_wts = copy.deepcopy(model.state_dict())
    best_acc = 0.0

    for epoch in range(num_epochs):
        print('Epoch {}/{}'.format(epoch, num_epochs - 1))
        print('-' * 10)

        # 每个epoch都有训练和验证阶段
        for phase in ['train', 'val']:
            if phase == 'train':
                model.train()  # 设置模型为训练模式
            else:
                model.eval()   # 设置模型为验证模式

            running_loss = 0.0
            running_corrects = 0

            # 迭代数据
            for inputs, labels in dataloaders[phase]:
                inputs = inputs.to(device)
                labels = labels.to(device)

                # 零参数梯度
                optimizer.zero_grad()

                # 前向传播
                # 训练模式下追踪历史，验证模式下不追踪历史
                with torch.set_grad_enabled(phase == 'train'):
                    features = pretrained_model(inputs)
                    outputs = model(features)

                    _, preds = torch.max(outputs, 1)
                    loss = criterion(outputs, labels)

                    # 训练阶段进行反向传播和优化
                    if phase == 'train':
                        loss.backward()
                        optimizer.step()

                # 统计损失和正确预测数量
                running_loss += loss.item() * inputs.size(0)
                running_corrects += torch.sum(preds == labels.data)

            # 更新学习率
            if phase == 'train':
                scheduler.step()

            # 计算训练和验证阶段的平均损失和准确率
            epoch_loss = running_loss / dataset_sizes[phase]
            epoch_acc = running_corrects.double() / dataset_sizes[phase]

            print('{} Loss: {:.4f} Acc: {:.4f}'.format(phase, epoch_loss, epoch_acc))

            # 深度复制模型
            if phase == 'val' and epoch_acc > best_acc:
                best_acc = epoch_acc
                best_model_wts = copy.deepcopy(model.state_dict())

        print()

    time_elapsed = time.time() - since
    print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed % 60))
    print('Best val Acc: {:4f}'.format(best_acc))

    # 加载最佳模型权重
    model.load_state_dict(best_model_wts)
    return model

创建模型实例

model_ft = OpenMAX(num_classes)
model_ft = model_ft.to(device)

定义优化器和学习率调整策略

optimizer_ft = optim.Adam(model_ft.parameters(), lr=learning_rate)
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)

训练模型

model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler, num_epochs)

保存模型

torch.save(model_ft.state_dict(), os.path.join(save_dir, 'model.pth'))

总结

本项目实现了基于OpenMAX的细粒度分类模型，该模型能有效地拒绝未知类别。代码使用了PyTorch框架，并提供了完整的训练过程。用户可以根据自己的需求修改参数和数据集进行训练。