IEEE - Aerospace and Electronic Systems - November 2019 - 47

automatically extract the relevant features from a set of
radar measurements or radar simulations that are stored in
the KB. The extracted features can then be used by the
KAP to classify the target.
Deep learning has become the preferred method for
many pattern recognition applications, such as segmentation, speech recognition, and face recognition since the
quantum leap in image recognition performance realized
with the AlexNet convolutional neural network (CNN) in
the ImageNet 2012 challenge [4]. Despite the major
achievements of deep learning techniques, such as CNNs
in the past few years, it is not obvious that the success of
deep learning in the commercial domain can be replicated
in the military domain. Large labeled datasets, which are
the key to the success of most commercial applications of
deep learning, are often not available in military applications. In addition, the cost of decision errors in the military
domain is typically much higher than in the commercial
applications. This makes it mandatory to achieve a robust
performance with a low error rate. Furthermore, military
ATR algorithms should not only be accurate, but their
behavior should also be predictable to gain the trust of
military commanders.
One of the first applications of deep learning for ATR
using synthetic aperture radar imagery was presented by
Morgan [5]. Since then many papers on the application in
radar of CNNs and other types of deep neural networks
have been published, including the application of deep
learning for target classification, and human gait and gesture recognition using micro-Doppler spectrograms [6]-[9].
This paper investigates the potential of deep learning
techniques for the classification of mini-UAVs using microDoppler spectrograms in the context of cognitive radar. The
paper also presents preliminary results on the use of deep
learning in the preprocessing and training process of a classifier. This paper is organized as follows. "CLASSIFICATION
OF MINI-UAVS" shows how deep neural networks can be
applied for the classification of mini-UAVs using measured
or simulated sets of micro-Doppler spectrograms that are
stored in the long-term memory (or knowledge base) of a
cognitive radar. "DETECTION OF UNKNOWN TARGET
NOVEMBER 2019

CLASSES" describes how the spectrograms of target classes
that are not represented in the training set, i.e., targets
unknown by the cognitive radar, can be detected with
deep neural networks. The detection of an unknown target
can provide a trigger for the radar scheduler to collect
micro-Doppler spectrograms of this target for inclusion in
the training set, i.e., the knowledge base. "DENOISING OF
SPECTROGRAMS" shows how a trained neural network
can be used to denoise spectrograms. This denoising process
may enable a cognitive radar to recognize targets at a longer
range. "ADVERSARIAL TRAINING FOR SPECTROGRAM GENERATION" describes the generation of new
spectrograms for training a classifier using generative adversarial networks. Finally, in "CONCLUSION," conclusions
are drawn with respect to the application of deep learning for
classification of mini-UAVs with cognitive radar.

CLASSIFICATION OF MINI-UAVS
The commercial availability of compact electronics and
advanced open source software has led to a proliferation
of mini-UAVs that can be used for many different purposes including criminal, terrorist, and military activities.
Owing to their relatively slow speed and small size, they
are hard to distinguish from natural targets, such as birds.
Recently, radar signal processing techniques using features extracted from spectrograms and cepstrograms have
been developed to discriminate birds and mini-UAVs
[10], [11]. The recognition of different classes of miniUAVs is, however, a more difficult problem due to the
overlapping features and rapidly changing characteristics
of mini-UAVs. Therefore, an enhanced capability for
radars is needed to recognize the type of mini-UAV for
threat evaluation and assignment of countermeasures.

MINI-UAV MEASUREMENTS
To investigate the potential of deep learning techniques
for the recognition of mini-UAVs, radar measurements of
five different types of mini-UAVs have been acquired

IEEE A&E SYSTEMS MAGAZINE

IEEE - Aerospace and Electronic Systems - November 2019

Table of Contents for the Digital Edition of IEEE - Aerospace and Electronic Systems - November 2019