【摘要】 针对当前广泛应用的网络流量加密技术在一定程度上对网络安全产生了严重影响的问题，基于深度学习提出了一种实时网络流量分类模型。所提方法对Inception-ResNet进行了轻量化改进，并结合加性裕度的Softmax对分类模型的损失函数进行改进；除此之外，采用通道剪枝技术，进一步对模型进行轻量化改进，并使用特征融合的在线蒸馏算法对模型进行训练。在公开数据集上的实验结果表明：所提方法能够对恶意流量实现精确分类，且对13种应用程序的平均分类准确率达到了99.23%，具有较好的细粒度分类效果，相较于其它对比模型具有显著优势。更多还原

【Abstract】 To address the significant impact of widely deployed encrypted network traffic on cybersecurity, a real-time network traffic classification model based on deep learning was proposed. Lightweight modifications to the Inception-ResNet architecture were implemented, and the loss function of the classification model was enhanced by incorporating additive margin Softmax. Additionally, channel pruning techniques were employed to further lightweight the model, while a feature fusion-based online distillation algorithm was utilized for model training. Experimental results on public datasets demonstrate that the proposed method achieves precise classification of malicious traffic, with an average classification accuracy of 99. 23% for 13 application types, exhibiting excellent fine-grained classification performance and significant advantages over other comparative models.更多还原