TY - JOUR
T1 - A Multi-Feature Radio-Frequency Framework for UAV Behavioral State Recognition and Prediction
AU - Mao, Runze
AU - Wu, Teng
AU - Wei, Shengjun
AU - Hu, Changzhen
N1 - Publisher Copyright:
© 2026 by the authors.
PY - 2026/9
Y1 - 2026/9
N2 - Existing RF-based UAV detection methods achieve high accuracy in identifying UAV presence and model type, yet they largely characterize UAVs through static signal attributes, offering limited insight into what an identified UAV is actually doing. This gap constrains their practical value for airspace monitoring and threat assessment. This paper presents a multi-feature RF-based framework for UAV behavioral state recognition and short-horizon behavior prediction, built upon a set of newly defined behavioral indicators, namely, spectral dynamics, signal-power-based motion trend, and communication density, integrated through a dedicated time-series modeling module. To support this study, we construct UAV-BehaviorRF, a new dataset with fine-grained behavioral annotations collected via a scripted multi-state flight protocol across eight UAV models. Experiments on UAV-BehaviorRF and the public DroneRFa dataset show that the proposed framework achieves accurate behavioral state recognition and reliable state-transition prediction, while remaining robust under interference and real-world conditions and maintaining real-time processing suitable for resource-constrained deployment.
AB - Existing RF-based UAV detection methods achieve high accuracy in identifying UAV presence and model type, yet they largely characterize UAVs through static signal attributes, offering limited insight into what an identified UAV is actually doing. This gap constrains their practical value for airspace monitoring and threat assessment. This paper presents a multi-feature RF-based framework for UAV behavioral state recognition and short-horizon behavior prediction, built upon a set of newly defined behavioral indicators, namely, spectral dynamics, signal-power-based motion trend, and communication density, integrated through a dedicated time-series modeling module. To support this study, we construct UAV-BehaviorRF, a new dataset with fine-grained behavioral annotations collected via a scripted multi-state flight protocol across eight UAV models. Experiments on UAV-BehaviorRF and the public DroneRFa dataset show that the proposed framework achieves accurate behavioral state recognition and reliable state-transition prediction, while remaining robust under interference and real-world conditions and maintaining real-time processing suitable for resource-constrained deployment.
KW - anomaly detection
KW - behavior prediction
KW - behavioral state recognition
KW - multi-feature extraction
KW - radio-frequency (RF) signal
KW - unmanned aerial vehicle (UAV)
UR - https://www.scopus.com/pages/publications/105051586498
U2 - 10.3390/drones10090668
DO - 10.3390/drones10090668
M3 - Article
AN - SCOPUS:105051586498
SN - 2504-446X
VL - 10
JO - Drones
JF - Drones
IS - 9
M1 - 668
ER -