V3I11P31

Fixed-Wing versus Multirotor UAV Recognition from Radar Track Kinematics Using a Jump-Markov Multiple-Model Estimator

Bui Tien Anh1*

Abstract

Recognising whether a small unmanned aerial vehicle (UAV) is of fixed-wing or multirotor type is a prerequisite for threat assessment and countermeasure selection in counter-UAS surveillance, yet many deployed surveillance radars provide only plot-level kinematics without micro-Doppler signatures. Falileev and Shatovkin (2020) proposed recognising the UAV type from track kinematics alone, exploiting the fact that only rotary-wing platforms can hover for a sustained period: a three-mode jump-Markov system (manoeuvre, non-manoeuvre, hover) is estimated with a second-order generalised pseudo-Bayesian (GPB2) filter, and a rotary-wing declaration is issued when the hover mode is estimated on n consecutive scans within the same azimuth sector.

This paper restates that algorithm in a reproducible form, analyses its transition probability matrices (TPMs), and evaluates it by Monte Carlo simulation over 1,300 synthetic radar tracks at a 3 s update period.

We show that the published TPMs are not row-stochastic and, once normalised, have identical rows, i.e. they describe a memoryless mode process that cannot accumulate evidence over scans. In our simulations the published parameterisation never declared a multirotor (0/600 tracks, measurement noise 1–25 m). Replacing the TPM with a diagonally dominant, persistence-aware matrix yields 100% recognition of multirotors at every noise level with a median latency of 9–12 s after hover onset and at most 1% false declarations on nominal fixed-wing tracks, against a median latency of 6–63 s for a filtered speed-threshold baseline. For low-airspeed fixed-wing aircraft flying into strong headwinds, a horizontal-velocity gate on the mode-conditioned estimate reduces false declarations from 38% to 8% (3-of-3 rule) at a cost of 3 s median latency. We discuss the latency–false-alarm trade-off governed by n and the fundamental limits of kinematics-only recognition.

Keywords:

unmanned aerial vehicle; target recognition; jump-Markov system; multiple-model estimation; GPB2; hover detection; counter-UAS; radar tracking.