Uncrewed aerial vehicles (UAVs) are becoming essential in various sectors, such as commercial delivery, agricultural monitoring, and disaster response. Despite their benefits, the rapid adoption of UAVs poses substantial security challenges, especially in drone network intrusion detection. Traditional intrusion detection datasets often suffer from limitations like small sample sizes and uneven distribution, undermining the effectiveness of intrusion detection systems. Moreover, conventional machine learning (ML) approaches generally require extensive, well-labeled datasets that are expensive and labor intensive to produce. To overcome these challenges, we introduce a generative adversarial network (GAN) model designed to enhance and balance the limited datasets available for drone networks. This model significantly improves data quality and quantity, thus optimizing the training process for intrusion detection models. Furthermore, we propose a Human-in-the-Loop (HITL) ML framework that integrates human expertise to guide the learning process and mitigate the costs of labeling. Our comprehensive evaluation demonstrates that the combined application of the GAN model and the HITL framework significantly outperforms traditional baseline models. This approach not only achieves an intrusion detection accuracy of up to 99% across various experimental datasets but also dramatically reduces the requirement for large amounts of labeled data by up to 98%, providing a cost-effective solution for enhancing UAV network security.

Drones, also called unmanned aerial vehicles (UAV), have become essential for surveillance tasks like warning systems, wildfire monitoring, and precision agriculture. To increase efficiency and energy savings, drone fleets are now prioritized over individual units. However, coordinating these fleets poses a challenge. Existing methods generally use a centralized control system for coordination and identification of events. In this paper, we introduce a cooperative and autonomous flocking method for drone piloting. This method mimics the behaviors of flocks of birds using a network coordination protocol and wireless communications. The foundation of this approach is the BOID model, which is based on three rules: separation, alignment and cohesion. Here, we improve this model by integrating detection and relative localization capabilities, resulting in the formation of dynamic and adaptable herds. The experiments carried out in simulation of our technique prove its ability to improve the performance of data exchanges and make routing more reliable, thus potentially improving the use of drones in various industries and needs.

Drone swarm coordination, inspired by natural swarm behaviors such as bird flocking, has traditionally hinged on rule-based methodologies, with Craig Reynolds' Boid algorithm as a seminal reference. These rule-based strategies, although functional in many contexts, fail to capture the intricate adaptive learning mechanisms inherent to birds navigating dynamic environments. Recognizing this limitation, this paper introduces a multi-agent reinforcement learning (MARL) approach to Boid modeling for drone swarms, its more authentic emulation of the nuanced learning processes witnessed in avian species. Our methodology leverages reinforcement learning algorithms to train individual drones, enabling them to autonomously make decisions based on their local environment and the overall swarm's state. Distinct from traditional rule-based models, our MARL-driven drones continually optimize their collective and individual behaviors, echoing the adaptability of natural flocks. This adaptability, intrinsically woven into the fabric of our model, offers heightened proficiency in navigating complex, mutable environments, bringing the simulation closer to the organic flocking dynamics observed in nature. Empirical evaluations showcase that our MARL Boid not only surpass traditional rule-based models in cohesion, alignment, and separation but also excel in adaptability to environmental variations. Moreover, our model is able to capture the flocking behavior quite effectively and show robustness against external perturbations.

The emergence of Unmanned Aerial Vehicle (UAV) networks has brought about intricate security challenges, especially in the domain of intrusion detection. At the same time, traditional machine learning algorithms for network security are becoming progressively inadequate, particularly when confronted with adversarial attacks and advanced persistent threats. Adding to the complexity, real-world UAV data streams are relentless, extensive, and demand substantial storage, making the storage of all data almost unfeasible. Furthermore, a significant gap in research lies in the lack of universally recognized datasets specifically designed for UAV network intrusion detection. Taking into account the aforementioned factors, this paper introduces an innovative approach for real-time intrusion detection. This approach harnesses Human-in-the-Loop Machine Learning (HITL-ML) to integrate human expertise directly into the machine learning process. By doing so, it enhances the system's capacity to adapt to evolving threats. Additionally, we introduce Generative Adversarial Networks (GANs) to synthetically generate data that mimics authentic network intrusions, thereby addressing the issue of limited dataset availability. This integration substantially enhances detection accuracy and reduces false positives, signifying a remarkable leap forward in contrast to conventional detection systems.

Vehicle ad-hoc networks represent a promising technology aimed at providing more efficient, safer, and partially autonomous transportation through vehicle-to-vehicle and infrastructure communications. The success of these networks heavily relies on the integrity and correctness of transmitted information. One of the most significant threats to communication is Denial of Service attacks, which can disrupt the system by overloading hardware modules. To effectively identify these attacks, we propose a modified version of the online K-Means algorithm. In our method, we aim to mitigate the impact of outliers on each adjusted centroid by incorporating an outlier detection method. In this paper, we compare the performance of the model by incorporating 4 different outlier detection methods in terms of execution time, accuracy and required buffer sizes. Additionally, we address a common oversight in accuracy calculations when using buffers for streaming data. Our study highlights the impact of data loss on accuracy. We also suggest several techniques to enhance the model in detecting DoS Attacks in terms of accuracy, precision and F1-score.

This paper introduces a novel real-time intrusion detection approach for UAV networks, addressing current security challenges. We leverage Generative Adversarial Networks (GANs) to generate synthetic data and implement a stream-based active learning process for prompt detection and response. Our method merges GANs with active learning, involving human experts to navigate the dynamic nature of UAV data and improve detection accuracy. Our results show that this system outperforms traditional intrusion detection methods, promising a new direction for future security frameworks and their application across various networked systems.