Single Shot Radar Target Detection and Localization using Deep Neural Network

Present day airborne radars have to perform multiple tasks like detection, tracking, and identification of targets in the presence of advanced EW countermeasures like RF jamming, and must be able to separate true targets from noise and clutter conditions caused by complex physical conditions, such as mountainous terrain, and environmental conditions. Although conventional signal processing methods for target detection have achieved a great level of maturity, they still continue to perform poorly under low SNR conditions and model imperfections due to statistical variations. Meanwhile, performing target detection using neural network-based methods has also been gathering increased attention. We show that various Convolutional Neural Network(CNN) and Recurrent Neural Network(RNN) based techniques improve target detection at low SNR conditions. The CNN-based architecture proposed in this paper performs single shot radar target detection on RD map, and localization of the target. Various enhancements to the technique is also proposed. The proposed method shows improvement over conventional and state-of-the-art deep learning based methods. Furthermore, we evaluate the proposed method on real airborne data and prove its effectiveness.