Deep Learning for Classification of Mini-UAVs Using Micro-Doppler Spectrograms in Cognitive Radar

Target recognition using micro-Doppler spectrograms measured with radar has many benefits when compared with optical target recognition due to the all-weather and day-andnight capability of radar. Micro-Doppler spectrograms are, however, difficult to interpret by radar operators. Consequently, there is a need for automatic target recognition methods for radar. Cognitive radars that learn to extract relevant features from measured or simulated micro-Doppler spectrograms stored in long-term memory, offer the potential to automatically recognize targets. In this paper, the ability of various deep learning techniques, such as convolutional neural networks and recurrent neural networks for the classification of mini-UAVs using micro-Doppler spectrograms is investigated. In addition, the detection of spectrograms from targets that are not known by the cognitive radar, denoising of spectrograms, and the generation of spectrograms for underrepresented target classes with generative adversarial networks is explored.