A Method of CNN Traffic Classification Based on Sppnet

Abstract

Nowadays, CNN widely used in network traffic classification. The traditional model of CNN only can be sent with the fixed traffic dataset in network traffic classification. But for the traffic dataset in network, that model must lead to a certain degree loss of the dataset by truncated or discarded. To solve this defect, a new CNN traffic classification model based on sppnet (spatial pyramid pooling) is proposed in this paper. Based on the CNN model of the LeNet-5, in the pooling layer before the fully connected layer, the new model is replaced the max-pooling to the spatial pyramid pooling which can realize the network traffic with indefinite length dataset. Through a series of experiments, the model has achieved certain achievement, and reducing the impact of human factors on traffic classification.