Drones Classification by the Use of a Multifunctional Radar and Micro-Doppler Analysis

The classification of targets by the use of radars has received great interest in recent years, in particular in defence and military applications, in which the development of sensor systems that are able to identify and classify threatening targets is a mandatory requirement. In the specific case of drones, several classification techniques have already been proposed and, up to now, the most effective technique was considered to be micro-Doppler analysis used in conjunction with machine learning tools. The micro-Doppler signatures of targets are usually represented in the form of the spectrogram, that is a time-frequency diagram that is obtained by performing a short-time Fourier transform (STFT) on the radar return signal. Moreover, frequently it is possible to extract useful information that can also be used in the classification task from the spectrogram of a target. The main aim of the paper is comparing different ways to exploit the drone's micro-Doppler analysis on different stages of a multifunctional radar. Three different classification approaches are compared: classic spectrogram-based classification; spectrum-based classification in which the received signal from the target is picked up after the moving target detector (MTD); and features-based classification, in which the received signal from the target undergoes the detection step after the MTD, after which discriminating features are extracted and used as input to the classifier. To compare the three approaches, a theoretical model for the radar return signal of different types of drone and aerial target is developed, validated by comparison with real recorded data, and used to simulate the targets. Results show that the third approach (features-based) not only has better performance than the others but also is the one that requires less modification and less processing power in a modern multifunctional radar because it reuses most of the processing facility already present.