A Multi-Scale Network Intrusion Detection Method Based on CNN and Transformer

Addressing the challenges in feature extraction for network intrusion detection due to vast high-dimensional traffic data, a novel approach integrating the Variational Gaussian Model (VGM) with a Convolutional Neural Network-Transformer (referred to as CNN-Transformer, or CTA-IDS) has been introduced, termed CTA-IDS. This system leverages VGM and OneHot encoding to preprocess the initial dataset, simplifying intricate features into more manageable components. Subsequently, a Convolutional Neural Network-Transformer fusion network designed for multi-scale feature selection, known as CTA-Net, is deployed to streamline network complexity. The efficacy of this approach is substantiated through experimental validation. The findings from these experiments indicate: 1) VGM-processed data significantly boosts detection precision; 2) When juxtaposed with three alternative convolutional neural networks, CTA-IDS markedly improves detection accuracy.