【摘要】 针对传统方法依靠人工提取特征、现有基于深度学习的方法在多分类上性能不够高等问题，提出一种基于稠密连接卷积网络的加密流量分类方法，利用DenseBlock中各层卷积稠密连接的结构，减轻梯度消失、加强特征传递，以提高分类性能。在公开数据集“ISCX VPN-nonVPN”上进行实验，结果表明该方法对不同种类的加密流量有更好的分类效果，准确率达到98.56%,F1值达到98.55%,相比于基于一维卷积神经网络模型和ResNet模型的方法，准确率分别提升了8.88百分点和6.54百分点，F1值分别提升了8.86百分点和6.64百分点。更多还原

【Abstract】 Aimed at the problems that traditional methods rely on manual feature extraction and the existing deep learning-based methods are not high enough for multi-classification, a densely connected convolutional network-based encryption traffic classification method is proposed. The densely connected structure of each layer of convolution in DenseBlock was used to reduce the disappearance of gradients and enhance feature transfer to improve classification performance. Experiments were conducted on the public data set "ISCX VPN-nonVPN". The results show that this method has a better classification effect on different types of encrypted traffic, with an accuracy of 98.56% and an F1 Score of 98.55%. Compared with the methods based on the 1D-CNN and the ResNet model, the accuracy is increased by 8.88 and 6.54 percentage points, and the F1 Score is increased by 8.86 and 6.64 percentage points, respectively.更多还原