Intrusion Detection System for MIL-STD-1553 Based on Convolutional Neural Networks With Binary Images and Adaptive Quantization

Abstract

This letter proposes an Intrusion Detection System (IDS) for the MIL-STD-1553 serial bus protocol, which is used in the aerospace systems. This letter proposes a novel encoding scheme to transform all the traffic data of MIL-STD-1553 including header, payload and time of packet transmission to binary images, which are given as an input to a Convolutional Neural Network (CNN). The encoding scheme is based on a quantization parameter $Q_{b}$ , which must be tuned to support the optimal attack detection performance of the algorithm. Then, this letter proposes a pre-processing adaptive step before the application of CNN to select the optimal value of $Q_{b}$ . The proposed approach is applied on a recently published cybersecurity data set of MIL-STD-1553 traffic, where it achieves a detection accuracy of 99.31%.