individual
What campus are you from?
Daytona Beach
Authors' Class Standing
Kristiyan Stefanov, Senior
Lead Presenter's Name
Kristiyan Stefanov
Faculty Mentor Name
Miroslav Stefanov
Abstract
This paper presents an intelligent hybrid system for detecting and responding to cyberattacks targeting Unmanned Aerial Vehicles (UAVs). Existing UAV defense systems rely heavily on centralized processing and cannot effectively detect new or unknown attacks in real time. To address this gap, the proposed architecture integrates two artificial intelligence models: a supervised Multilayer Perceptron (MLP) for classifying known threats and an unsupervised Autoencoder (AE) for detecting behavioral anomalies. Both models operate onboard using edge AI hardware, allowing autonomous processing without ground connectivity. The system analyzes telemetry, GPS, IMU, and network control data to detect attacks such as GPS spoofing, RF hijacking, and Denial-of-Service (DoS). It also includes a response logic module that triggers autonomous defensive actions, including command filtering, safe hover mode, and IMU-only navigation when GPS spoofing is suspected. Evaluation on a dataset of 42 real and simulated records containing over 300,000 entries shows a classification accuracy of 96.7%, F1-scores above 0.98, and anomaly detection precision of 0.941, with response times below 0.5 seconds. The results confirm that the hybrid AI system improves UAV resilience and autonomy by enabling real-time, onboard detection and response to both known and novel cyber threats.
Did this research project receive funding support from the Office of Undergraduate Research.
No
Intelligent Detection and Response System for Attacks on Unmanned Aerial Vehicles Based on Hybrid AI Models, MLP and Autoencoder
This paper presents an intelligent hybrid system for detecting and responding to cyberattacks targeting Unmanned Aerial Vehicles (UAVs). Existing UAV defense systems rely heavily on centralized processing and cannot effectively detect new or unknown attacks in real time. To address this gap, the proposed architecture integrates two artificial intelligence models: a supervised Multilayer Perceptron (MLP) for classifying known threats and an unsupervised Autoencoder (AE) for detecting behavioral anomalies. Both models operate onboard using edge AI hardware, allowing autonomous processing without ground connectivity. The system analyzes telemetry, GPS, IMU, and network control data to detect attacks such as GPS spoofing, RF hijacking, and Denial-of-Service (DoS). It also includes a response logic module that triggers autonomous defensive actions, including command filtering, safe hover mode, and IMU-only navigation when GPS spoofing is suspected. Evaluation on a dataset of 42 real and simulated records containing over 300,000 entries shows a classification accuracy of 96.7%, F1-scores above 0.98, and anomaly detection precision of 0.941, with response times below 0.5 seconds. The results confirm that the hybrid AI system improves UAV resilience and autonomy by enabling real-time, onboard detection and response to both known and novel cyber threats.