Research and Improvement of Encrypted Traffic Classification Based on Convolutional Neural Network
Yansen Zhou, Jianquan Cui · 2020
Aiming at the problems of low recognition rate and long training time of deep convolution neural network Alexnet in encrypted traffic classification, some improvement measures are put forward for the classical Alexnet network, mainly including the introduction of multi-scale convolution, deconvolution operation and batch standardization, which can extract more comprehensive features and reduce convolution kernel parameters. The performance of the improved Alexnet convolutional neural network is tested by encrypting the traffic dataset. The test results show that the recognition accuracy and precision of the improved model on the selected test set are 83.9% and 84% respectively, which are about 7.2% and 8% higher than those of the classical model. The improved Alex net model has a certain improvement in the performance of network encryption traffic classification recognition.