BLSAE-SNIDS: A Bi-LSTM sparse autoencoder framework for satellite network intrusion detection

Abstract

Due to disparities in tolerance, resource availability, and acquisition of labeled training data between satellite-terrestrial integrated networks (STINs) and terrestrial networks, the application of traditional terrestrial network intrusion detection techniques to satellite networks poses significant challenges. This paper presents a satellite network intrusion detection system named Bi-LSTM sparse selfencoder (BLSAE-SNIDS) to address this issue. Through the development of an innovative unsupervised training Bi-LSTM stacked self-encoder, BLSAE-SNIDS facilitates feature extraction from satellite network traffic, diminishes dimensionality, considerably reduces training and testing durations, and enhances the attack prediction accuracy of the classifier. To assess the efficacy of the proposed model, we conduct comprehensive experiments utilizing STIN and UNSW-NB15 datasets. The results obtained from the STIN dataset demonstrate that BLSAE-SNIDS achieves 99.99% accuracy with reduced computational and transmission overheads alongside enhanced flexibility. Furthermore, results from the UNSW-NB15 dataset exhibit BLSAE-SNIDS? proficiency in detecting various network intrusion attacks efficiently. These findings indicate that BLSAE-SNIDS suits general satellite security networks and offers a novel approach to designing security systems for polar satellite networks, thus exhibiting practical utility.