Secure Networking and Efficient Computing in Unmanned Aerial Vehicle Assisted Green IoT Networks

The integration of Unmanned Aerial Vehicles (UAVs) with Green Internet of Things (GIoT) provides a transformative approach to environmental monitoring, enhancing the flexibility, efficiency and sustainability in various applications such as smart cities, wildfire detection, military operations, and smart agriculture. UAVs, equipped with cameras, LiDAR, and radar, excel in providing remote sensing capability with high mobility and significant computational capacities. However, their effectiveness is limited by a lack of direct, in situ sensing. Conversely, GIoT network bring high scalability and sustainability through equipped in situ sensors, yet they struggle with limited energy, communication and computing resources. By combining UAVs with GIoT, the resultant system leverages the strengths of both: enhancing in situ data collection with UAVs' broad surveillance capabilities and GIoT's dense, ground truth data. In this study, we envision a UAV-assisted GIoT network where Energy Harvesting Nodes (EHNs) harvest renewable energy to sense and communicate within a GIoT network. UAVs function as mobile sink nodes to collect data from these EHNs and are equipped with cameras to implement computer vision models, aiding in intelligent route planning and enhancing operational efficiency. Due to the computational demands of deep learning models and the limited on-board processing capabilities of UAVs, there is a necessity to offload some computational tasks to an edge server. This task offloading allows the UAVs to run complex algorithms efficiently without compromising their performance. This dissertation develops secure networking and efficient computing solutions for UAV-assisted GIoT networks. It encompasses three research thrusts. In the first research thrust of this dissertation, we identify a new security vulnerability within GIoT networks, called ``malicious energy attacks''. These attacks exploit the energy-aware routing selection designed to prolong network life. By intentionally charging specific nodes, the energy attacker can manipulate routing paths and attract most data traffic through a compromised node, undermining information security. In this work, we develop intelligent energy attack methods utilizing Q-learning and Policy Gradient reinforcement learning algorithms. In the second thrust, we delve into the comprehensive analysis of the impacts of malicious energy attacks and the development of an ML-enabled security routing algorithm. This security routing approach aims to accurately detect the presence of energy attack and mitigate the impact of the energy attack accordingly, ensuring the integrity and reliability of data transmission. The third research thrust of the dissertation addresses the challenges of long latencies and high network bandwidth demands in UAV-assisted IoT networks, particularly concerning task offloading in mobile edge computing. To improve real-time performance, we propose a novel feature map compression method based on statistical analysis of feature map data. This method is rigorously tested through extensive experiments, which demonstrate significant enhancements in computing efficiency, thereby optimizing the performance of UAV-assisted IoT networks.

Grantor dc:publisher

University of Alabama Libraries

Year dc:date.issued

2024

Author dc:creator

• Li, Long

Advisor dc:contributor.advisor

• Pu, Lina

Contributors dc:contributor

• Gong, Jiaqi
• Hong, Xiaoyan
• Bangalore, Purushotham
• Luo, Yu

• IoT
• Machine Learning
• Mobile Edge Computing
• Wireless Sensor Network

Statement dc:rights

• All rights reserved by the author unless otherwise indicated.

Language dc:language.iso

en_US, English

Dc Identifier Other

1067710

OAI identifier oai:identifier

oai:ir.ua.edu:123456789/14361