EAS CSIS Doctoral Dissertation Defense by Mengjie Jia
Advisor: Dr. Jiawei Yuan, Department of Computer & Information Science
Committee members:
- Dr. Gokhan Kul, Department of Computer and Information Science
- Dr. Long Jiao, Department of Computer and Information Science
- Dr. Ruolin Zhou, Department of Electrical & Computer Engineering
Abstract
Unmanned Aerial Vehicles (UAVs) are widely used in civilian and military applications such as surveillance, reconnaissance, and traffic monitoring. These missions rely on object tracking to detect and follow targets in real time. Recent advances in artificial intelligence have enhanced UAV tracking performance through the integration of deep learning–based visual trackers. However, these trackers also raise security concerns due to the vulnerability of deep learning models to adversarial attacks. Malicious perturbations on video frames can mislead trackers and result in mission failures. Since the assumption of attack-free operating environments is not always guaranteed, UAV tracking must remain reliable under attack to ensure mission success. This dissertation develops robust defense solutions for building secure UAV tracking systems in adversarial environments. For Siamese-based trackers, a pluggable dual-level reconstruction approach is proposed to mitigate adversarial impacts at both the feature and decision levels without modifying tracker architectures. For transformer-based trackers, a two-stage defense solution is proposed, combining adaptive search region adjustment with input reconstruction to counter diverse attacks while maintaining accuracy on clean data. Comprehensive evaluations on UAV tracking benchmarks demonstrate that the proposed solutions can effectively restore performance and enhance the robustness of UAV tracking systems under adversarial conditions. Real-world tests on UAV platforms further validate the efficiency and feasibility of the solutions.
For further information, please contact Dr. Jiawei Yuan.
Dion 311
Dr. Jiawei Yuan
X8457
jyuan@umassd.edu