一种基于深度特征融合的网络流量分类方法

A network traffic classification method based on deep feature fusion

随着网络应用的持续发展,识别特定的流或者应用程序有着重要的作用。由于机器学习方法对特征选择较为苛刻,因而,具有自动特征提取的深度学习算法的优势就突显出来了。但现有深度学习方法大多是对网络流量的原始字节进行处理和分类,而原始字节包含较多的冗余和混淆信息。针对此,提出了一种基于深度特征融合的流量分类方法。该方法对原始的统计特征进行融合,并将融合后的特征转化为灰度图像,应用卷积神经网络对转换后的灰度图像进行分类,达到对加密流量进行分类的目的。在两个真实数据集上进行实验验证,分类准确率达到了92%~99.89%。与文献方法相比,在网络流量粗粒度和细粒度分类上都取得了更好的结果。

As network applications have been increasingly developed,identifying specific traffic or applications has become critical.Since machine learning methods are more dependent on feature selection,the advantages of deep learning algorithms with automatic feature extraction have become prominent.However,most deep learning methods process and classify the original bytes of network traffic that contain considerable redundant and confusing information.Therefore,a traffic classification method based on deep feature fusion is proposed.In this method,the original statistical features are fused,and the fused features are transformed into grayscale images.The convolution neural network is used to classify the transformed grayscale images,and the purpose of classifying encrypted traffic is achieved.Verification experiments are carried out on two data sets that glean data from the real world.The results show that the classification accuracy of the proposed method reaches 92—99.89%,and that compared with classic methods,the proposed method obtain better coarse⁃grained and fine⁃grained classification of network traffic.