A Method for Anomaly Detection of Unmanned Aerial Vehicles (UAVs)

Abstract

The aim of this study is to detect anomalies and improve the safety of UAVs by analysing UAV flight data temporally and spatially. Flight parameters and sensor data were recorded on a UAV platform built using PIXHAWK3 autopilot hardware. The data analysed with Python and related libraries were converted into a weighted graph with 'networkx' and visualised with 2D graphics. In the flights performed in ten different scenarios, a total of 105 anomalies were detected, especially in the first flight and in the 1130-1160 m altitude range. The temporal and spatial graph neural network method stands out as a critical tool for flight safety by effectively identifying anomalies in UAV flights. ©2024 European Union.