Robust Drone Classification Using Two-Stage Decision Trees and Results from SESAR SAFIR Trials

Jahangir, Mohammed; Ahmad, Bashar I.; Baker, Chris

doi:10.1109/radar42522.2020.9114870

Cited by 36 publications

(34 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…There have been several studies addressing the classification of drones in radar systems due to the increased demand for this technology [7]. A collection of these papers utilize a kinematic feature set that includes interpretable attributes such as velocity, acceleration, and curvature [5,8,9]. These studies have demonstrated how kinematic features can be employed as part of machine learning algorithms.…”

Section: B Related Work and Contributionsmentioning

confidence: 99%

“…This paper proposes a new feature that accounts for the shape and interaction of micro-Doppler components for different classes of targets while maintaining a very explainable nature. This feature has been designed to be adopted as part of interpretable classifiers, such as those proposed in [8] and offers a unique separation between drones and other targets. In order to demonstrate the robust nature of this feature, a substantial dataset has been collected of diverse experimental signatures containing data from fixed wing and rotary wing drones from three different locations and for flights up to 2.6 km radar range.…”

Section: B Related Work and Contributionsmentioning

confidence: 99%

“…The track features are derived both from the kinematic features and also contain features relating to the Doppler spectrum. Earlier work has only considered the number of Doppler harmonics that can be attributed to a tracked target [6,8,15]. The Doppler feature extraction is performed using a dedicated module that detects the number of micro-Doppler components corresponding to a given track.…”

Section: Staring Radar System Overviewmentioning

confidence: 99%

“…Although there are some additional Doppler echoes present, these are random and are most likely the result of the Doppler return from other nearby birds. A common approach for incorporating micro-Doppler components in an explainable drone classifier is to count the number of these components [6,8,15]. Drones typically exhibit four or more, whereas birds usually have only two components at most.…”

Section: Dronesmentioning

confidence: 99%

“…The flights in these recordings span many days of trials in two different locations out to a maximum range of 2.6 km. Flights were conducted using a DJI Inspires 1 and 2 as part of the SESAR CLASS [9] and SESAR SAFIR [8] consortiums trials program and as such, were recorded in a live setting without any optimization or conditioning of the surrounding environment. The 55 flights that are included in this dataset make up for 34% of the data, with other background targets such as birds filling the rest.…”

Section: A Evaluation Datasetmentioning

confidence: 99%

See 4 more Smart Citations

Use of Symmetrical Peak Extraction in Drone Micro-Doppler Classification for Staring Radar

Bennett

Jahangir

Fioranelli

et al. 2020

2020 IEEE Radar Conference (RadarConf20)

Self Cite

View full text Add to dashboard Cite

The commercialization of drones has granted the public with unprecedented access to unmanned aviation. As such, the detection, tracking, and classification of drones in radars have become an area in high demand to mitigate accidental or voluntary misuse of these platforms. This paper focuses on the classification of drone targets in a safety context where the concept of Explainable AI is of particular interest. Here, we propose a simple, yet effective, means to extract a salient symmetry feature from the micro-Doppler signatures of drone targets, arising from onboard rotary components. Most importantly, this approach maintains the explainable nature of the employed recognition algorithm, as the symmetry feature is directly related to the kinematics of the drones as the targets of interest. A large dataset collected from multiple locations with over 280 minutes of rotary and fixed wing drone flights has been collected and used to demonstrate the generalization capability of this approach.

show abstract

Section: B Related Work and Contributionsmentioning

confidence: 99%

Section: B Related Work and Contributionsmentioning

confidence: 99%

Section: Staring Radar System Overviewmentioning

confidence: 99%

Section: Dronesmentioning

confidence: 99%

Section: A Evaluation Datasetmentioning

confidence: 99%

See 3 more Smart Citations

Use of Symmetrical Peak Extraction in Drone Micro-Doppler Classification for Staring Radar

Bennett

Jahangir

Fioranelli

et al. 2020

2020 IEEE Radar Conference (RadarConf20)

Self Cite

View full text Add to dashboard Cite

show abstract

Intensive Review of Drones Detection and Tracking: Linear Kalman Filter Versus Nonlinear Regression, an Analysis Case

Zitar

Mohsen

Seghrouchni

et al. 2023

Arch Computat Methods Eng

View full text Add to dashboard Cite

SNR‐dependent drone classification using convolutional neural networks

Dale

Baker

Antoniou

et al. 2021

IET Radar Sonar & Navi

View full text Add to dashboard Cite

Radar sensing offers a method of achieving 24-h all-weather drone surveillance, but in order to be maximally effective, systems need to be able to discriminate between birds and drones. This work examines drone-bird classification performance as a function of signal to noise ratio (SNR). Classification at low SNR values is necessary in order to classify drones with a small radar cross-section (RCS), as well as to facilitate reliable classification at longer ranges. To investigate the relationship between classification performance and SNR, Gaussian noise is added to an experimentally obtained dataset of radar spectrograms. Classification is performed by convolutional neural networks (CNNs). It is shown that for the data available classification accuracy drops with falling SNR, as might be expected for any given CNN. The degree to which performance degrades with reduced SNR is presented. It is further shown that simpler network architectures are more robust to noise. Finally, it is demonstrated that data augmentation can be used as a means of enhancing classification accuracy at lower SNR values. Bayesian optimisation is used to find the optimal augmentation hyperparameters and overall, classification accuracies of 92% are achieved at low SNR.This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

show abstract

Robust Drone Classification Using Two-Stage Decision Trees and Results from SESAR SAFIR Trials

Cited by 36 publications

References 15 publications

Use of Symmetrical Peak Extraction in Drone Micro-Doppler Classification for Staring Radar

Use of Symmetrical Peak Extraction in Drone Micro-Doppler Classification for Staring Radar

Intensive Review of Drones Detection and Tracking: Linear Kalman Filter Versus Nonlinear Regression, an Analysis Case

SNR‐dependent drone classification using convolutional neural networks

Contact Info

Product

Resources

About