Date of Award

8-2023

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Civil Engineering

Committee Chair/Advisor

Dr. Mashrur Chowdhury

Committee Member

Dr. Yao Wang

Committee Member

Dr. Feng Luo

Committee Member

Dr. Long Cheng

Committee Member

Dr. Sakib Mahmud Khan

Abstract

The rapid advancement of communication and artificial intelligence technologies is propelling the development of connected and autonomous vehicles (CAVs), revolutionizing the transportation landscape. However, increased connectivity and automation also present heightened potential for cyber threats. Recently, the emergence of generative neural networks (NNs) has unveiled a myriad of opportunities for complementing CAV applications, including generative NN-based cybersecurity measures to protect the CAVs in a transportation cyber-physical system (TCPS) from known and unknown cyberattacks. The goal of this dissertation is to explore the utility of the generative NNs for devising cyberattack detection and mitigation strategies for CAVs. To this end, the author developed (i) a hybrid quantum-classical restricted Boltzmann machine (RBM)-based framework for in-vehicle network intrusion detection for connected vehicles and (ii) a generative adversarial network (GAN)-based defense method for the traffic sign classification system within the perception module of autonomous vehicles. The author evaluated the hybrid quantum-classical RBM-based intrusion detection framework on three separate real-world Fuzzy attack datasets and compared its performance with a similar but classical-only approach (i.e., a classical computer-based data preprocessing and RBM training). The results showed that the hybrid quantum-classical RBM-based intrusion detection framework achieved an average intrusion detection accuracy of 98%, whereas the classical-only approach achieved an average accuracy of 90%. For the second study, the author evaluated the GAN-based adversarial defense method for traffic sign classification against different white-box adversarial attacks, such as the fast gradient sign method, the DeepFool, the Carlini and Wagner, and the projected gradient descent attacks. The author compared the performance of the GAN-based defense method with several traditional benchmark defense methods, such as Gaussian augmentation, JPEG compression, feature squeezing, and spatial smoothing. The findings indicated that the GAN-based adversarial defense method for traffic sign classification outperformed all the benchmark defense methods under all the white-box adversarial attacks the author considered for evaluation. Thus, the contribution of this dissertation lies in utilizing the generative ability of existing generative NNs to develop novel high-performing cyberattack detection and mitigation strategies that are feasible to deploy in CAVs in a TCPS environment.

Author ORCID Identifier

0000-0001-7326-3694

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.