Enhancing Cybersecurity of Automotive Ethernet Networks with Deep Learning-based Intrusion Detection Systems

The growing number of attack surfaces in vehicles, driven by increased connectivity and the demand for automotive Ethernet, a high-bandwidth in-vehicle network technology, emphasizes the need for effective security mechanisms. This dissertation proposes two deep learning-based intrusion detection systems (IDSs) for identifying cyberattacks in automotive Ethernet networks. The first proposal features an IDS based on a multi-criteria optimized convolutional neural network, designed to enhance detection accuracy, speed, and storage efficiency simultaneously. The second proposal introduces a multi-stage deep learning-based IDS, where the initial stage prioritizes fast detection while the second stage focuses on achieving more accurate results and classifying attacks. The main results of this dissertation comprehend the publication of a paper in SBRC 2023 and another in the Ad Hoc Networks journal.