A multiscale intrusion detection system based on pyramid depthwise separable convolution neural network

Abstract

To solve the feature extraction problem in network intrusion detection, which is caused by large-scale high-dimensional traffic data, we propose a method based on variational Gaussian model (VGM) and one-dimensional Pyramid Depthwise Separable Convolution (PyDSC) neural network, called PyDSC-IDS. PyDSC-IDS uses VGM and OneHot encode technologies to preprocess the original dataset and decompose the complex feature into several simple ones. Pyramid convolution (PyConv) is selected for the processed multi-scale features and Depthwise Separable Convolution (DSC) is added to PyConv to reduce its network complexity. Finally, we verify the availability of the proposed method through experiments. The experimental results show that: 1) Data processed by VGM can effectively improve the detection accuracy; 2) Compared with the other three convolutional neural networks, PyDSC can significantly improve the detection accuracy at the cost of a slight increase in complexity, and PyDSC can effectively reduce the complexity of the PyConv.