Abstract:Unmanned Aerial Vehicles (UAVs), commonly referred to as drones, have determined numerous applications across industries, ranging from aerial surveillance to package shipping. As drones are used in vital operations, ensuring cyberattacks and anomalies protect them from attackers is now a big challenge. This research study presents a complete approach to enhancing drone safety by integrating multi-sensor anomaly detection and superior machine learning techniques. The proposed methodology capitalizes on the rich sensor suite embedded in present-day drones, encompassing GPS receivers, accelerometers, gyroscopes, cameras, communication modules, and more. Leveraging an array of sensors in drones, our technique detects abnormal drone behavior indicative of unauthorized access, GPS spoofing, communication jamming, and malicious activities. By extracting features from sensor records, we develop a robust anomaly detection framework using the "uav attack dataset" able to identify deviations from normal flight patterns, communication signals, and environmental interactions. Central to our methodology is the utilization of machine learning algorithms. These algorithms are skilled on labeled datasets containing numerous flight eventualities, each normal and hostile, together with the ones discovered inside the "uav attack dataset". The obtained results are eventually evaluated using rigorous performance metrics to quantify their effectiveness in distinguishing genuine anomalies from benign variations. The findings of our study underscore the capacity of multi-sensor anomaly detection for drones. By harnessing the power of machine learning and sensor fusion, we exhibit the ability to hit upon attacks at an early level, mitigating capability harm and permitting rapid responses. This study contributes now not only to the field of drone safety but also to the broader panorama of self-sustaining systems protection, highlighting the importance of adaptive and proactive protection mechanisms. Results show that an accuracy of 99% with AUC of 100$ was achieved when all the sensors were used.

Identifying safety–critical concerns in unmanned aerial vehicle software platforms with SALIENT

Automated Identification and Qualitative Characterization of Safety Concerns Reported in UAV Software Platforms

Automated Unearthing of Dangerous Issue Reports.

Automatic detection of safety requests in web and mobile applications using natural language processing techniques

Stability and Accuracy-Aware Learning for Task Offloading in UAV-MEC-Assisted Smart Farms

MAVFI: An End-to-End Fault Analysis Framework with Anomaly Detection and Recovery for Micro Aerial Vehicles

Enhancing Drone Security Through Multi-Sensor Anomaly Detection and Machine Learning

Assuring Intelligent Systems: Contingency Management for UAS

Enhancing Situational Awareness of Helicopter Pilots in Unmanned Aerial Vehicle-Congested Environments Using an Airborne Visual Artificial Intelligence Approach

Secure and Privacy-Preserving Intrusion Detection and Prevention in the Internet of Unmanned Aerial Vehicles

Safety Concerns and Mitigation Approaches Regarding the Use of Deep Learning in Safety-Critical Perception Tasks

Understanding Social Perception, Interactions, and Safety Aspects of Sidewalk Delivery Robots Using Sentiment Analysis

Uncovering Safety Risks of Large Language Models through Concept Activation Vector

Scope Compliance Uncertainty Estimate

SAFETY-J: Evaluating Safety with Critique

System Safety Engineering for Social and Ethical ML Risks: A Case Study

Towards Engineering Fair and Equitable Software Systems for Managing Low-Altitude Airspace Authorizations

Identification, Categorisation and Gaps of Safety Indicators for U-Space

Robust and Explainable Semi-Supervised Deep Learning Model for Anomaly Detection in Aviation

Security-Informed Safety Analysis of Autonomous Transport Systems Considering AI-Powered Cyberattacks and Protection

SafeLLM: Domain-Specific Safety Monitoring for Large Language Models: A Case Study of Offshore Wind Maintenance